Dirichlet Series

Dirichlet Series
John E. McCarthy
Copyright c 2018 John E. McCarthy
All rights reserved.
Contents
Preface v
Notation vii
Chapter 1. Introduction 1
1.1. Exercises 10
1.2. Notes 10
Chapter 2. The Prime Number Theorem 11
2.1. Statement of the Prime number theorem 11
2.2. Proof of the Prime number theorem 12
2.3. Historical Notes 21
Chapter 3. Convergence of Dirichlet Series 23
Chapter 4. Perron’s and Schnee’s formulae 29
4.1. Notes 35
Chapter 5. Abscissae of uniform and bounded convergence 37
5.1. Uniform Convergence 37
5.2. The Bohr correspondence 42
5.3. Bohnenblust-Hille Theorem 43
5.4. Notes 53
Chapter 6. Hilbert Spaces of Dirichlet Series 55
6.1. Beurling’s problem: The statement 55
6.2. Reciprocals of Dirichlet Series 59
6.3. Kronecker’s Theorem 60
6.4. Power series in infinitely many variables 61
6.5. Besicovitch’s Theorem 63
6.6. The spaces Hw2 66
6.7. Multiplier algebras of H2 and Hw2 69
6.8. Cyclic Vectors 74
6.9. Exercises 74
6.10. Notes 75
Chapter 7. Characters 77
iii
iv CONTENTS
7.1. Vertical Limits 77

7.2. Helson’s Theorem 82
7.3. Dirichlet’s theorem on primes in arithmetic progressions 86
7.4. Exercises 93
7.5. Notes 93
Chapter 8. Zero Sets 95
Chapter 9. Interpolating Sequences 99
9.1. Interpolating Sequences for Multiplier algebras 99
9.2. Interpolating sequences in Hilbert spaces 105
9.3. The Pick property 107
9.4. Sampling sequences 109
9.5. Exercises 110
9.6. Notes 110
Chapter 10. Composition operators 113
Chapter 11. Appendix 115
11.1. Multi-index Notation 115
11.2. Schwarz-Pick lemma on the polydisk 115
11.3. Reproducing kernel Hilbert spaces 117
11.4. Multiplier Algebras 118
Bibliography 121
Index 125
Preface
In 2005, I taught a graduate course on Dirichlet series at Washing-

ton University. One of the students in the course, David Opěla, took
notes and TeX’ed them up. We planned to turn these notes into a
book, but the project stalled.
In 2015, I taught the course again, and revised the notes. I still
intend to write a proper book, eventually, but until then I decided to
make the notes available to anybody who is interested. The notes are
not complete, and in particular lack a lot of references to recent papers.
Dirichlet series have been studied since the 19th century, but as
individual functions. Henry Helson in 1969 [Hel69] had the idea of
studying function spaces of Dirichlet series, but this idea did not really
take off until the landmark paper [HLS97] of Hedenmalm, Lindqvist
and Seip that introduced a Hilbert space of Dirichlet series that is anal-
ogous to the Hardy space on the unit disk. This space, and variations
of it, has been intensively studied, and the results are of great interest.
I would like to thank all the students who took part in the courses,
and my two Ph.D. students, Brian Maurizi and Meredith Sargent, who
did research on Dirichlet series. I would especially like to thank David
Opěla for his work in rendering the original course notes into a leg-
ible draft. I would also like to thank the National Science Founda-
tion, that partially supported me during the entire long genesis of this
project, with grants DMS 0501079, DMS 0966845, DMS 1300280, DMS
1565243.
v
Notation
N = {0, 1, 2, 3, . . . }
N+ = {1, 2, 3, 4, . . . }
Z = integers
Q = rationals
R = reals
C = complex numbers
P = {2, 3, 5, 7, . . . } = {p1 , p2 , p3 , p4 , . . . }
Pk = {p1 , p2 , . . . , pk }
Nk = {n ∈ N+ : all prime factors of n lie in Pk }
s = σ + it, s ∈ C, σ, t ∈ R
Ωρ = {s ∈ C; Re s > ρ}
π(x) = # of primes ≤ x
µ(n) = Möbius function
d(k) = number of divisors of k
dj (k) = number of ways to factor k into exactly j factors
φ(n) = PEuler totient function
Φ(s) = p∈P log ps
p
P
Θ(x) = p≤x log p
σc = abscissa of convergence
σa = abscissa of absolute convergence
σ1 = max(0, σc )
σu = abscissa of uniform convergence
σb = abscissa of bounded convergence
F (x) summatory function
Z T
− normalized integral
−T
εn = Rademacher sequence
E = Expectation
T = torus
tl tl
z r(n)P:= z1t1 . . . zP t1
l , where n = p1 · · · pl
B : P an z r(n) 7→ P an n−s
Q : an n−s 7→ an z r(n)
∞
T = infinite torus
vii
viii NOTATION
√
β(x) = P2 sin(πx) P
H2 = { ∞ n=1 an n
−s
: 2
n |an | < ∞}
Mult (X ) = {ϕ : ϕf ∈ X , ∀f ∈ X }
Mϕ : f 7→ ϕf
D∞ = infinite polydisk
E(ε, f ) ε-translation numbers of f
Hw2 = weighted space of Dirichlet series
Hw2 =weighted space of power series
P∞ −s
7→ n∈NK an n−s
P
QK : n=1 an n
ρ = Haar measure on T∞
`2 (G) = Hilbert space of square-summable functions on the group G
Xq = Dirichlet characters modulo q
L(s, χ) = Dirichlet L series
R θ+1 1
p
H∞ (Ω1/2 ) = {g ∈ Hol (Ω1/2 ) : [supθ∈R supσ>1/2 θ |g(σ + it)|p dt] p < ∞}
h RT i1/p
1 p
kf kHp = limT →∞ 2T −T
|f (it)| dt
4 The left-hand side is less than or equal to a constant times the right-hand side
≈ Each side is 4 the other side
ρA (x, y) = sup{|φ(y)k : φ(x) = 0, kφk ≤ 1}
H∞ = H ∞ (Ω0 ) ∩ D
E : f 7→ hf, gi i
gi = kλi /kkλi k
CHAPTER 1
Introduction
A Dirichlet series is a series of the form

X∞
an n−s =: f (s), s ∈ C.
n=1
The most famous example is the Riemann zeta function

∞
X 1
ζ(s) = s
.
n=1
n
Notation 1.1. By long-standing tradition, the complex variable

in a Dirichlet series is denoted by s, and it is written as
s = σ + it.
We shall always use σ for <(s) and t for =(s).
Note 1.2. The Dirichlet series for ζ(s) converges if σ > 1; in fact,
it converges absolutely for such s, since
|n−s | = |e−(σ+it) log n | = |e−(σ+it) log n | = n−σ .
Also, if σ ≤ 0 or 0 < s ≤ 1, the series diverges, in the first case because
the terms do not tend to zero, in the second by comparison with the
harmonic series.
Remark 1.3. Consider the power series ∞ n
P
n=1 z ; it converges to
1
1−z
, but only in the open unit disk. Nonetheless, it determines the
1
analytic function f (z) = 1−z everywhere, since it has a unique ana-
lytic continuation to C \ {1}. The Riemann zeta function can also be
analytically continued outside of the region where it is defined by the
series.
For this continuation, it can be shown that ζ(−2n) = 0, for all n ∈
+
N and that there are no other zeros outside of the strip 0 ≤ Re s ≤ 1.
The Riemann hypothesis, proposed by Bernhard Riemann in 1859, is
one of the most famous unanswered conjectures in mathematics. It
states that all the zeros other than the even negative integers have real
part equal to 12 .
1
2 1. INTRODUCTION
We shall prove in Theorem 2.19 that the zeta function has no zeroes
on the line {<s = 1}.
The importance of the Riemann zeta function and the Riemann

hypothesis lies in their intimate connection with prime numbers and
their distribution. On the simplest level, this can be explained by the
Euler Product formula below.
Recall that an infinite product ∞
Q
n=1 an is said to converge, if the
partial products tend to a non-zero finite number P (or if one of the an ’s is
zero). This is equivalent to the requirement that ∞ n=1 log an converges
(or an = 0, for some n ∈ N+ ). See e.g. [Gam01, XIII.3].
Notation 1.4. We shall let P denote the set of primes, and when
convenient we shall write
P = {p1 , p2 , p3 , p4 , . . . } = {2, 3, 5, 7, . . . }
to label the primes in increasing order. We shall let Pk denote the first
k primes.
Theorem 1.5. (Euler Product formula) For σ > 1,
Y −1 ∞
1 X
1− s = ζ(s) = n−s .
p∈P
p n=1
Formal proof:
Y −1 −1 −1 −1

1 1 1 1
1− s = 1− 1− s 1− s ...
p∈P
p 2s 3 5

1 1 1
= 1+ + + + ... ×
2s 22s 23s

1 1 1
1+ + + + ... ...
3s 32s 33s
If we formally multiply out this infinite product, we can only obtain

a non-zero product by choosing 1 from all but finitely many brackets.
This product will be qr1 s qr21s ...qrk s = n1s . For each n ∈ N+ , the term
1 2 k
1
ns
will appear exactly once, by the existence and uniqueness of prime
factoring.
1. INTRODUCTION 3
For a rigorous proof assume that Re s > 1, and fix k ∈ N+ . Then

Y −1 Y
1 1 1 1
1− s = 1 + s + 2s + 3s + . . .
p∈Pk
p p∈Pk
p p p
X 1
= s
, (1.6)
r1 rk n
n=p1 ...pk
where the last equality holds by a variation of the formal argument

above and convergence is not a problem, since we are multiplying
finitely many absolutely convergent series.
Using (1.6), we have, for Re s > 1,

−1
Y 1 X 1 X 1
ζ(s) − 1− s = ≤ → 0, as k → ∞.

p∈P
p ns n≥p
nσ
k
{n; pl |n, l>k} k+1
Thus the product converges to ζ(s).

To see that the limit is non-zero, we have

1
1 − (1 − ) ≤ 1
−1 1

ps pσ pσ − 1
2
≤ for p large.
pσ
Since σ > 1, this
P means that the infinite product converges absolutely,
1 −1
and therefore log(1 − ps ) converges absolutely.
Notation 1.7. We shall let Ωρ denote the open half-plane

Ωρ = {s : <(s) > ρ}.
Corollary 1.8. ζ(s) has no zeros in Ω1 .
Proof: For s ∈ Ω1 , ζ(s) is given by an absolutely convergent
product.
Thus, it can only be zero if one of the terms is zero. But
−1
1
1− ps
= 0 if and only if ps = 0, which never happens.
1
P
Theorem 1.9. = ∞.
p∈P p
Proof: Suppose not, then p∈P p1 converges. By the Taylor expan-

P
sion of log(1 − x), for x close enough to 0,

x
−x ≤ log(1 − x) ≤ − ,
2
4 1. INTRODUCTION

1
P
so we conclude that p∈P log 1 − p
also converges. Since

1 1
log 1 − < log 1 − σ ,
p p
for all σ > 1 and p ∈ P, we get

X 1
−∞ < log 1 −
p∈P
p

X 1
< lim log 1 − σ
σ→1+
p∈P
p
1
= − lim log
σ→1+ ζ(σ)
= −∞,
a contradiction.
The following discrete version of integration by parts is often useful
when working with Dirichlet series. In it, integrals are replacedRn by
sums, and derivatives by Rdifferences. (In the familiar formula m udv =
n
u(n)v(n) − u(m)v(m) − m vdu, we let u correspond to b, v to A and
thus dv to a.)
In fact, one can prove integration by parts for Riemann integrals
using the definition (via Riemann sums) and Lemma 1.10.
PnLemma 1.10. (Abel’s Summation by parts formula) Let An =

k=1 ak , then
n
X n−1
X
ak bk = An bn − Am−1 bm + Ak (bk − bk+1 ).
k=m k=m
Proof: Since ak = Ak − Ak−1 , we have
n
X n
X
ak b k = [Ak − Ak−1 ]bk
k=m k=m
Xn n
X
= Ak bk − Ak−1 bk
k=m k=m
n−1
X
= Ak [bk − bk+1 ] − Am−1 bm + An bn .
k=m

1. INTRODUCTION 5
Notation 1.11. For x > 0, we let π(x) denote the number of

primes less than or equal to x.
The prime number theorem (see Chapter 2) is an estimate of how
big π(n) is for large n. We can use the Euler product formula to relate
π and the Riemann zeta function.
Theorem 1.12. For σ > 1,
Z ∞
π(x)
log ζ(s) = s dx .
2 x(xs − 1)
Proof: In the following calculation we use the fact that [π(k) −
π(k − 1)]Pis equal to 1 if k is a prime, and 0 if k is composite; the
equality nk=1 [π(k) − π(k − 1)] = π(n); and summation by parts.

X 1
log ζ(s) = − log 1 − s
p∈P
p
∞
X 1
= − [π(k) − π(k − 1)] log 1 − s
k=2
k
L
X 1
= − lim [π(k) − π(k − 1)] log 1 − s
L→∞
k=2
k
(L−1
X 1 1
= − lim π(k) log 1 − s − log 1 −
L→∞
k=2
k (k + 1)s

1 1
+ π(1) log 1 − s − π(L) log 1 − s
2 L
The penultimate term vanishes, since π(1) = 0. As for the last term,
the trivial bound π(L) ≤ L gives

π(L) log 1 − 1 ≤ L · L−σ → 0 as L → ∞.

Ls
d
We let L → ∞, and use the fact that dx log(1 − x1s ) = xs+1s −x , to get:
∞
X 1 1
log ζ(s) = − π(k) log 1 − s − log 1 −
k=2
k (k + 1)s
∞ Z k+1
X −s
= − π(k) s+1
dx
k=2 k x −x
Z ∞
π(x)
= s s+1
dx .
2 x −x

6 1. INTRODUCTION
Notation 1.13. The Möbius function is helpful when working

with the Riemann zeta function. It is given as follows:

1, n = 1,

µ(n) = (−1)k , if n is the product of k distinct primes,

0, otherwise.
Its values for the first few positive integer are in the table below:
n 1 2 3 4 5 6 7 8 9 10 11 12
µ(n) 1 -1 -1 0 -1 1 -1 0 0 1 -1 0
∞
1 X
= µ(n)n−s .
ζ(s) n=1
Proof: We only present a formal proof — convergence can be

checked in the same way as was done for the Euler product formula.
1 Y 1

= 1− s
ζ(s) p∈P
p

1 1 1
= 1− s 1− s 1 − s ...
X2 3X 5
−s −s −s
= 1− p + p q − ...
p∈P p,q∈P, p6=q
∞
X µ(n)
=
n=1
ns

P∞
It is obvious that the Dirichlet series n=1 an n−s converges (con-
verges
P∞ absolutely, respectively) for all s ∈ Ωρ if and only if the series
−ρ −s
(a
n=1 n n )n converges (conv. abs., resp.) for all s ∈ Ω0 . This
ability to translate the Dirichlet series horizontally often allows one to
simplify calculations. (It is analogous to working with power series and
assuming the center is at 0). The proof of the proposition below is a
typical example of this.
The following “uniqueness-of-coefficients” theorem will be used fre-
quently.
Proposition 1.15. Suppose that ∞ −s
P
n=1 an n converges absolutely
to f (s) in some half-plane Ωρ and f (s) ≡ 0 in Ωρ . Then an = 0 for all
n ∈ N+ .
1. INTRODUCTION 7
P Proof: As remarked, we may assume that ρ < 0, so in particular,

|an | < ∞. Suppose all the an ’s are not 0, and let n0 be the smallest
natural number such that an0 6= 0.
Claim: limσ→∞ f (σ)nσ0 = an0 .
To prove the claim note that

X
0 ≤ nσ0 an n−σ

n>n0
X n σ
0
≤ |an |
n>n0
n
σ X
n0
≤ |an |,
n0 + 1 n>n
0
P
and the last term tends to 0 as σ → ∞, since |an | converges. As
X
f (σ)nσ0 = an0 + nσ0 an n−σ ,
n>n0
the claim is proved.

The proof is also finished, because the limit in the claim is obviously
0, a contradiction.
Recall that the Cauchy product formula for the product of power
series states that
∞
!
X X X X
an z n bm z m = an bk−n z k ,
k=0 0≤n≤k
if at least one of the sums on the left-hand side converges absolutely.

The Dirichlet series analogue below involves the sum over all divisors of
a given integer. The multiplicative structure of the natural numbers is
far more complex than their additive structure. Indeed, as an additive
semigroup N+ is singly generated, while as a multiplicative semigroup
it is not finitely generated — the smallest set of generators is P. This is
one of the reasons why the theory of Dirichlet series is more complicated
than the theory of power series. Now, we state the Dirichlet series
analogue of the Cauchy product formula. The proof is immediate.
Theorem 1.16. Assume that ∞ −s
and ∞ −s
P P
n=1 an n m=1 bm m con-
verge absolutely. Then
 
∞
! ∞ ! ∞
X X X X
an n−s bm m−s =  an bk/n  k −s ,
n=1 m=1 k=1 n|k
with absolute convergence.

8 1. INTRODUCTION
Corollary 1.17. For σ > 1,

∞
X
ζ 2 (s) = d(k)k −s ,
k=1
where d(k) denotes the number of divisors of k. More generally,

∞
X
j
ζ (s) = dj (k)k −s ,
k=1
where dj (k) denotes the number of ways to factor k into exactly j fac-
tors. Here, 1 is allowed to be a factor and two factorings that differ
only by the order of the factors are considered to be distinct.
Proof: We shall prove the first formula. Using Theorem 1.16, we
have, for σ > 1,
∞
! ∞ !
X X
ζ 2 (s) = n−s m−s
n=1 m=1
 
∞
X X
=  1 k −s
k=1 n|k
∞
X
= d(k)k −s .
k=1
The proof of the second formula is analogous.

1
The formula for ζ(s) implies the following identity for the Möbius
function. (It can also be proved directly.)
P
Corollary 1.18. n|k µ(n) = 0, for all k ≥ 2.
Proof: For σ > 1, write
1 = ζ(s)ζ −1 (s)
∞
! ∞ !
X 1 X µ(n)
=
m=1
ms n=1
ns
 
∞
X X 1
=  µ(n) s .
k=1
k
n|k
Comparing the coefficients of the outer-most Dirichlet series completes

the proof.
1. INTRODUCTION 9
Proposition 1.19. (Möbius inversion formula) Let f, g be

functions on N+ . If
X X
g(q) = f (n), then f (q) = µ(q/p)g(d).
n|q d|q
Proof:
X X X
µ(q/p)g(d) = µ(q/p) f (n)
d|q d|q n|d
 
X X
=  µ(q/d) f (n)
q q
d|q, d | n 
n|q
X X
=  µ(s) f (n)
q
n|q s| n
= f (q),
since, by the preceding corollary, the bracket is non-zero only when
q/n = 1.
Definition 1.20. The Euler totient function φ(n) is defined as

#{1 ≤ k ≤ n; gcd(n, k) = 1}.
Clearly, φ(p) = p − 1, iff p is a prime. In fact, one can express φ(n)
in terms of the prime factors of n.
Lemma 1.21. If n = q1r1 . . . qkrk with rj > 0, then
k
Y 1
φ(n) = n 1− .
j=1
qj

Proof: First, note that gcd(n, m) 6= 1, if and only if, qj m, for
some 1 ≤ j ≤ k. Consider the uniform probability distribution on
{1, . . . , n}. Let Ej be the event that qj divides a randomly chosen
number in {1, . . . , n}. For any l that divides n, there are exactly n/l
numbers in {1, . . . , n} divisible by l. Thus, the events {Ej }kj=1 are
independent and hence so are their complements. Hence, φ(n)/n, the
probability that a randomly chosen number is not divisible by any qj ,
is equal toQ the product of the probabilities that it is not divisible by
qj , that is j (1 − 1/qj ).

∞
ζ(s − 1) X φ(n)
= s
.
ζ(s) n=1
n
10 1. INTRODUCTION
Proof: Again, we will only prove it formally, since turning it into a

rigorous proof is routine, but renders the proof harder to read. By the
Euler product formula, we have

1
ζ(s − 1) Y 1 − ps
=
ζ(s) p∈P 1 − ps−1
1
Y p2

1 p
= 1− s 1 + s + 2s + . . .
p∈P
p p p
p2 p2

Y p 1 p
= 1 + s + 2s + . . . − s + 2s + 3s + . . .
p∈P
p p p p p
p2

Y 1 p
= 1+ 1− + + ...
p∈P
p ps p2s
X∞
= an n−s ,
n=1
where
k k
Y 1 r
Y 1
an = 1− qj j = n 1− = φ(n),
j=1
qj j=1
qj
for n = q1r1 . . . qkrk .
1.1. Exercises
1. Prove that if χ : N+ → T∪{0} is a quasi-character, which means
χ(mn) = χ(m)χ(n), the same argument that proved the Euler product
formula (Theorem 1.5) shows that
∞ Y
X χ(n) 1
= .
n=1
n−s p∈P
1 − χ(p)p−s
1.2. Notes
For a thorough treatment of the Riemann zeta function, see [Tit86].
The material in this chapter comes from the first few pages of Titch-
marsh’s magisterial book.
CHAPTER 2
The Prime Number Theorem
2.1. Statement of the Prime number theorem

We have defined π(n) to be the number of primes less than or equal
to n. Euclid’s proof that there are an infinite number of primes says
that limn→∞ π(n) = ∞; but how fast does it grow? By Theorem 1.9
and Abel’s summation by parts formula we know
X1
∞ =
p∈P
p
∞
X 1
= [π(n) − π(n − 1)]
n=2
n
∞
X 1
≈ π(n) ,
n=2
n2
so π(n) cannot be O(nα ) for any α < 1.

Gauss conjectured that
x
π(x) ∼ , (2.1)
log x
where the asymptotic symbol ∼ in (2.1) means that the ratio of the
quantities on either side tends to 1 as x → ∞. Tchebyshev proved that
x x
.93 ≤ π(x) ≤ 1.1
log x log x
for x large, and also showed that if
π(x)
lim
x→∞ x/ log x
exists, it must be 1. The full prime number theorem was proved in

1896 by de la Vallée Poussin and Hadamard.
Theorem 2.2. [Prime Number Theorem]
x
π(x) ∼ .
log x
11
12 2. THE PRIME NUMBER THEOREM
Looking at some examples, we see that


π(106 ) = 78, 498
 π(106 )
10 6 =⇒ 106
≈ 1.08
≈ 72, 382 log 106
log(106 )

and 
π(109 ) = 50, 847, 478
 π(109 )
109 =⇒ 109
≈ 1.05
≈ 48, 254, 942 log 109
log(109 )

Definition 2.3. For s ∈ Ω1 , we define

X log p
Φ(s) := s
.
p∈P
p
It is easy to see that this Dirichlet series converges absolutely in Ω1 .

Definition 2.4. For x ∈ R, define
X
Θ(x) := log p.
p∈P, p≤x
The key to proving the Prime number theorem is establishing the

estimate Θ(x) ∼ x (Proposition 2.27).
Say more here?
2.2. Proof of the Prime number theorem

We will now prove the Prime number theorem in a series of steps.
Lemma 2.5.
Θ(x)
Θ(x) = O(x) as x → ∞, i.e., lim sup < ∞.
x→∞ x
Proof: Note that

2n Y
≥ p.
n n<p≤2n, p∈P
Indeed, the LHS is a positive integer that is divisible by the RHS. Note
that we have not yet proved that there are any primes between n and
2n, so the RHS may be an empty product (we interpret empty products
as having the value 1).
Thus,
2n Y
≥ p = eΘ(2n)−Θ(n) .
n n<p≤2n
2.2. PROOF OF THE PRIME NUMBER THEOREM 13
Now, by the binomial theorem,

2n 2n 2n 2n 2n
2 = (1 + 1) = + ··· + + ··· + ,
0 n 2n
Thus
2n 2n n log 4 2n
2 ≥ =⇒ e ≥ ≥ eΘ(2n)−Θ(n) ,
n n
and consequently
Θ(2n) − Θ(n) ≤ n log 4.
For x ∈ R, x ≥ 1, we have
Θ(2x) − Θ(x) ≤ Θ(b2xc) − Θ(bxc)
≤ Θ(2 bxc + 1) − Θ(bxc)
≤ log(b2xc + 1) + Θ(b2xc) − Θ(bxc)
≤ cx.
x x
Now fix x and choose n ∈ N such that 2n+1
≤ 1 ≤ 2n
. Then, by
telescoping,
n
X x x
Θ(x) − Θ(1) = Θ( j
) − Θ( j+1 )
j=0
2 2
n
X x
≤ c
j=0
2j+1
= cx.
Since Θ(1) = 0, we conclude that

Θ(x) = O(x) (2.6)
2
Recall that Φ(s) = p∈P log p 1
P P
ps
. Since p∈P p = ∞ (Theorem 1.9)
we conclude that Φ(s) has a pole at 1.
1
Lemma 2.7. The function Φ(s) − s−1
is holomorphic in Ω1 .
Proof: In Ω1 ,
Y −1 Y
1 1
ζ(s) = 1− s =
p∈P
p p∈P
1 − ps
By logarithmic differentiation we obtain

ζ(s) X ∂
∂s
(1− p−s )
= −
ζ 0 (s) p∈P
1 − p−s
X 1
= − (p−s log p)
p∈P
1 − p−s
X log p
= − (2.8)
p∈P
ps − 1
Now
1 1 1
= s+ s s (2.9)
−1 p ps
p (p − 1)
Combining (2.8) and (2.9), we obtain, for s ∈ Ω1 ,
−ζ 0 (s) X log p X log p
= + (2.10)
ζ(s) p∈P
ps p∈P
ps (ps − 1)
Note that we can rearrange the terms since the series converge abso-
lutely in Ω1 . Thus, for s ∈ Ω1 ,
−ζ 0 (s) X log p
= Φ(s) + s (ps − 1)
(2.11)
ζ(s) p∈P
p
The second term on the RHS defines an analytic function in Ω1/2 as

the series converges there absolutely. Thus in Ω1/2 , any information
0
about the analyticity of −ζζ(s)(s) translates into the analyticity of Φ(s).
1
The function ζ(s) has a pole at 1 with residue 1 and so ζ(s) − s−1
is analytic near 1, and consequently, ζ 0 (s) + (s−1)1
2 is analytic near 1.
1
ζ 0 (s) − ζ 0 (s)
(s−1)2 1
Thus ζ(s)
− 1 = ζ(s)
− (s−1)
is analytic near 1.
s−1
Thus
ζ 0 (s) X log p
Φ(s) = − (2.12)
ζ(s) p∈P
ps (ps − 1)
is holomorphic in Ω1/2 ∩ {s : ζ(s) 6= 0}.
It remains to prove that ζ(s) 6= 0 if Re s ≥ 1.
Definition 2.13. The von Mangoldt function Λ : N0 → R, is
defined as (
log p, if m = pk ,
Λ(m) = (2.14)
0, else.
Proposition 2.15. For s ∈ Ω1
−ζ 0 (s) X Λ(n) X log p

= = (2.16)
ζ(s) n≥2
ns p∈P
ps − 1
holds.
1 −1
Q
Proof: We have ζ(s) = p∈P (1 − ps
) and thus
ζ 0 (s) X p−s
= − log p
ζ(s) p∈P
1 − p1s
X log p
= −
p∈P
ps − 1
which proves that the first and last term in the statement of the propo-
sition are equal. For Re s > 1, k1/ps k < 1, so the first equality above
yields
ζ 0 (s) X 1 1
= − (log p)p−s (1 + s
+ 2s + . . . )
ζ(s) p∈P
p p
∞
XX
= − log p(pk )−s .
p∈P k=1
This double summation goes over exactly those numbers n = pk for

which Λ(n) does not vanish and thus, for s ∈ Ω1 ,
∞
ζ 0 (s) X X X Λ(n)
− = log p(pk )−s = (2.17)
ζ(s) p∈P k=1 n≥2
ns
Lemma 2.18. Let x0 ∈ R and assume F is holomorphic in a neigh-

borhood of x0 , F (x0 ) = 0 and F 6= 0. Then there exists ε > 0 such
that
0
F (x)
Re >0
F (x)
for x ∈ (x0 − ε, x0 + ε).

Proof: Write F (x) = ak (s − x0 )k + ak+1 (s − x0 )k+1 + . . . where

k > 0. Then
F 0 (x) kak (x − x0 )k−1 + (k + 1)ak+1 (x − x0 )k + . . .
=
F (x) ak (x − x0 )k + ak+1 (x − x0 )k+1 + . . .
(k+1)ak+1
k+ ak
(x − x0 ) + . . .
= ak+1
(x − x0 ) + ak (x − x0 )2 + . . .
k
≈ > 0,
x − x0
for x ∈ (x0 , x0 + ε). 2
Theorem 2.19. The Riemann ζ function does not vanish on the
line {<(s) = 1}.
Proof: Suppose that ζ(1 + it0 ) = 0, for t0 ∈ R \ {0}. Define
F (s) := ζ 3 (s)ζ 4 (s + it0 )ζ(s + 2it0 ). (2.20)
At s = 1, we see that ζ 3 has a pole of order 3, and ζ 4 (s+it0 ) vanishes to
order 4, so F (1) = 0. Thus,
0 ina neighborhood of 1, F is holomorphic.
Using Lemma 2.18, Re FF (x) (x)
> 0 for x ∈ (1, 1 + ε). Computing
F 0 (x) ζ 0 (x) ζ 0 (x + it0 ) ζ 0 (x + 2it0 )

= 3 +4 +
F (x) ζ(x) ζ(x + it0 ) ζ(x + 2it0 )
X
Λ(n) −3n−x − 4n−x e−it0 log n − n−x e−2it0 log n ,

=
n≥2
thus,
F 0 (x) X
Re = −Λ(n)n−x [3 + 4 cos(t0 log n) + cos(2t0 log n)]
F (x) n≥2
X
= −Λ(n)n−x [2 + 4 cos(t0 log n) + 2 cos(t0 log n)]
n≥2
We observe that −Λ(n)n−x ≤ 0 for every n ≥ 2 while the term in the

square bracket is always non-negative, since it the square of
√
2[1 + cos(t0 log n)],
a contradiction with Lemma 2.18. 2
Lemma 2.21. Let f (t) : [0, ∞) → C be bounded and suppose that
Z ∞
g(s) = f (t)e−st dt (2.22)
0
R∞
extends to a holomorphic function in Ω0 . Then 0
f (t) dt exists and
equals g(0).
Proof: Let
Z T
gT (s) = f (t)e−st dt . (2.23)
0
Then gT is an entire function by Morera’s theorem, and gT (0) =

RT
0
f (t) dt . We want to show that limt→∞ gT (0) = g(0).
insert image around here
For R, δ > 0 let UR,δ := D(0, R) ∩ Ω−δ . For any R > 0 there
is δ > 0 such that g is holomorphic in UR,δ , since by hypothesis g is
holomorphic in a neighborhood of Ω0 . Let C := ∂UR,δ and C+ = C ∩Ω0
and C− = C \ Ω0 . By Cauchy’s theorem:
s2
Z
1 sT ds
g(0) − gT (0) = [g(s) − gt (s)] e 1+ 2 , (2.24)
2πi C R s
2
since est (1 + Rs 2 ) has value 1 at 0 and is
holomophic
everywhere in our
2
contour. Let h(s) := [g(s) − gt (s)] esT 1 + Rs 2 . In Ω0 , we have
∞
Z
−st

|g(s) − gT (s)| =
f (t)e dt
T
Z ∞
e−st

≤ M dt
ZT ∞
e−(Re
s)t

= M dt
T
1 −Re s T
= M e
Re s
Thus
Z Z
ds f (t)e−st dt

h(s) ≤

C+ s C+
e−Re sT
Z sT 2

e s
≤ M 1 + | ds |
C+ Re s s R2
For s ∈ C+ , we have |s| = R and so
s2 1 R 2 + s2 |s|2 + s

s+s
1+ 2 = 2
= 2
=
R s R s R s R2
Thus,
e−Re sT eRe sT 2Re s
Z Z
ds M
h(s) ≤ | ds |

C+ s 2π C+ Re s s R2
M
≤ πR
πR2
M
=
R
We conclude C+ h(s) ds
R
s
→ 0 as R → 0.
For C− , we will show that both
s2
Z
sT ds
I1 (T ) := |g(s)| e 1+ 2
C− R s
and
s2
Z
sT ds
I2 (T ) := |gT (s)| e 1+ 2
C− R s
tend to 0 as R tends to 0.
We start with I1 :
T
Z
−st

|gT (s)| =
f (t)e dt
0
Z T
≤ M e−(Re s)t
dt
0
Z T
≤ M e−(Re s)t
dt
−∞
M −Re st
= e
Re s
Therefore,
M e−(Re s)t sT s2 1
Z
I1 (T ) ≤ e 1+ 2 | ds |
C− |Re s| R s
and since gT is an entire function, we can integrate over the semicircle
Γ− instead of C− and use the same estimates as in Ω0 to get
M
I1 (T ) ≤ .
R
Now
s2 1 sT

Z
I2 (T ) = g(s) 1 + 2 e ds
C− R s
and the expression in square bracket is independent of T and holo-

morphic in a neighborhood of C− while esT → 0 as T → ∞. Using
dominated convergence theorem, we conclude that I2 → 0 as T → ∞.
Thus,
Z
ds
|g(0) − g(T )| ≤ h(s) + |I1 (T )| + |I2 (T )
C+ s
M M 2M
≤ + + I2 (T ) →
R R R
Taking the limit as R → ∞ implies that g(0) = limT →∞ gT (0) 2
R∞
Lemma 2.25. The integral 1 Θ(x)−x
x2
dx converges.
Proof: For Re s > 1,
X log p Z ∞ dΘ(x)
Φ(s) = s
=
p∈pri
p 1 dx
We are using the Stiltjes integral in the last expression because Θ(x)
is a step function.
We use integration by parts with u := x−s and dv := dΘ(x). Then
du = −sx−(s+1) dx and v(x) = Θ(x), giving
∞ Z ∞
−s Θ(x)
Φ(x) = x Θ(x) + s dx .

1 1 xs+1
The first term vanishes since Θ(x) = O(x) as x → ∞. We conclude
that Z ∞
Θ(x)
Φ(s) = s dx .
1 xs+1
Now let us use the substitution, x = et to get
Z ∞
Φ(s) = s Θ(et )e−ts dt .
0
We want apply Lemma 2.21 to f (t) := Θ(et )e−t −1 and g(s) = Θ(s+1)
s+1
−
1
s
. By Lemma 2.5, we get that f (t) is bounded and by Lemma 2.7, we
know that Θ(s+1)
s+1
− 1s is holomorphic in Ω0 . In order to apply Lemma
2.21, we need to check that g(s) is the Laplace transform of f (t).
We have
Z ∞ Z ∞
t −t −ts
Θ(e )e e dt = Θ(et )e−t(s+1) dt
0 0
and Z ∞
1
1e−ts dt =
0 s
and thus g(s) is the Laplace

R ∞ transform of f (t), and we can apply Lemma
2.21 to conclude that 0 f (t) dt exists.
Z ∞ Z ∞
Θ(et )e−t − 1 dt

f (t)dt =
0
Z0 ∞
1 dx
= Θ(x) − 1
1 x x
Z ∞
Θ(x) − x
= dx
1 x2
which concludes the proof. 2
Note 2.26. See [Fol99, p. 107] for information on integration by

parts in the context of the Stieltjes integrals.
Θ(x)
Proposition 2.27. limx→∞ x
= 1, that is, Θ(x) ∼ x.
Proof: We will proceed by contraction. There are two cases.

First assume that lim supx→∞ Θ(x)
x
> 1. Thus, there exists λ > 1
and a sequence {xn } with xn → ∞ such that Θ(xn ) > λxn . Then,
since Θ is non-decreasing,
Z λxn Z λxn
Θ(t) − t λxn − t
2
dt ≥ dt =: cλ .
xn t xn t2
We integrate the two pieces,
Z λxn
λxn 1 λxn
dt = λx n − =λ−1
t2 t xn

xn
and
Z λxn
dt
= log(λxn ) − log xn = log λ
xn t
to conclude that cλ = Rλ − 1 − log λ > 0 by a well-known inequality for
∞
log. This implies that 1 Θ(x)−x
x2
dx does not converge, a contradiction.
The second case is that lim inf x→∞ Θ(x)x
< 1, so there is λ < 1 and
Θ(xn )
a sequence {xn } with xn → ∞ and xn < λ. As before,
xn λxn
Θ(t) − t λxn − t
Z Z
dt ≤ dt = −cλ = −(λ − 1 − log λ) < 0
λxn t2 xn t2
and we reach a contraction as in the first case. 2
2.3. HISTORICAL NOTES 21
Proof of Theorem 2.2. We can estimate

X
Θ(x) = log p
p≤x
X
≤ log x
p≤x
= π(x) log x.
By Proposition 2.27, Θ(x) ∼ x and thus we have the bound
log x
lim sup π(x) ≥ 1.
x→∞ x
For the other bound, let ε > 0, and write
X
Θ(x) ≥ log p
x1−ε ≤p≤x
X
≥ log x1−ε
x1−ε ≤p≤x
= [π(x) − π(x1−ε )](1 − ε) log x

(1 − ε) log x π(x) + O(x1−ε )

=
where the last equality come from Lemma 2.5.
We have
1
π(x) log x ≤ Θ(x) + O(x1−ε log x)
1−ε
and hence
π(x) log x 1 Θ(x)
≤ + O(x−ε log x).
x 1−ε x
Using Proposition 2.27 again, we get
log x 1
lim sup π(x) ≤
x→∞ x 1−ε
for every ε > 0. Taking ε → 0 yields
log x
lim sup π(x) ≤ 1
x→∞ x
which concludes the proof. 2
2.3. Historical Notes

R x dt
The offset logarithmic integral function, Li(x) := 2 log t
satisfies
x
Li(x) ≈ log x ≈ π(x) but is a better approximation to π(x).
Gauss conjectured that π(n) ≤ Li(n). This was disproved by E.
Littlewood in 1914.
During the proof of the Prime Number Theorem, we used the fact
that ζ(s) does not vanish for Re s ≥ 1. More precise estimates showing
that the zeros of ζ(s) must lie “close to” the critical line {Re s = 1/2}
yield estimates on the error |π(x) − Li(x)|.
The Riemann hypothesis is equivalent to the error estimate
√
π(x) = Li(x) + O( x log x).
The best known estimate is of the error is
A(log x)3/5
−
π(x) = Li(x) + O(xe (log log x)1/5 ).
CHAPTER 3
Convergence of Dirichlet Series
We will now investigate convergence of Dirichlet series. Much of

the general theory holds for generalized Dirichlet series, that is, series
of the form
∞
X
an e−λn s .
n=1
An ordinary Dirichlet series corresponds, of course, to the case λn =

log n.
When dealing with a generalized Dirichlet series, we shall always
assume that λn is a strictly increasing sequence tending to infinity, and
that λ1 ≥ 0. Sometimes, an additional assumption is needed, such as
the Bohr condition, namely λn+1 −P λn ≥ c/n, for some c > 0.
Recall that for a power series ∞ n
n=1 an z there exists a (unique)
value R ∈ [0, ∞], called the radius of convergence, such that
(1) if |z| < R, then P∞ n
P
n=1 an z converges,
∞
(2) if |z| > R, then n=1 an z nPdiverges,
(3) for any r < R, the series ∞ n
n=1 an z converges uniformly and
absolutely in {|z| ≤ r} and the sum is bounded on this set,
(4) on the circle {|z| = R}, the behavior is more delicate.
As we shall see, the situation for Dirichlet series is more compli-
cated. In particular, compare the third point above with Proposi-
tion 3.10.
We start with a basic result.
Theorem 3.1. If the series ∞ −s

P
n=1 an n converges at some s0 ∈ C,
then, for every δ > 0, it converges uniformly in the sector {s : − π2 +δ <
arg(s − s0 ) < π2 − δ}.
P
Proof: P
As usual, we may assume s0 = 0, that is, n an converges.
Let rn := ∞ k=n+1 ak , and fix ε > 0. Then there exist n0 ∈ N such that
|rn | < ε for all n ≥ n0 . Using summation by parts, for s in the sector
23
24 3. CONVERGENCE OF DIRICHLET SERIES
and M, N > n0
N
X N
X
−s
an n = (rn−1 − rn )n−s
n=M n=M
N −1
X 1 1
= rn s
− s (3.2)
n=M
(n + 1) n
rM −1 rN
+ s
− s.
M N
The absolute values of the last two terms are bounded by ε, since their
numerators are bounded by ε while the denominators have absolute
value at least 1. To estimate (3.2), note that
Z n+1
1 1 −s
s
− s = dx ,
(n + 1) n n xs+1
so that
Z n+1
1 1 dx |s| 1 1
(n + 1)s − ns ≤ |s| = − . (3.3)

n |xs+1 | σ nσ (n + 1)σ
Thus the absolute value of (3.2) satisfies, for M, N > n0 ,

N −1 N −1
X 1 1 X |s| 1 1
rn − ≤ |rn | −
(n + 1)s ns σ nσ (n + 1)σ

n=M n=M
N −1
|s| X 1 1
≤ ε −
σ n=M nσ (n + 1)σ

|s| 1 1
≤ ε −
σ Mσ Nσ
≤ c(δ)ε, (3.4)
since |s| = |1/ cos(arg s)| ≤ 1/ cos π2 − δ =: c(δ). This proves that

σ
the series is uniformly Cauchy, and hence uniformly convergent.
P∞
Corollary 3.5. If n=1 an n−s converges at s0 ∈ C, then it con-
verges in Ωσ0 .
S
Proof: This follows from the inclusion Ωσ0 ⊂ δ>0 {s : arg |s−s0 | <
π
2
− δ}.
This implies that there exists a unique value σc ∈ [−∞, ∞] such
that the Dirichlet series converges to the right of it, and diverges to the
left of it.
3. CONVERGENCE OF DIRICHLET SERIES 25
Definition 3.6. The abscissa of convergence of the Dirichlet se-

ries ∞ −s
P
n=1 na n is the extended real number σc ∈ [−∞, ∞] with the
following properties
(1) if Re s > σc , then P∞ −s
P
n=1 an n converges,
∞ −s
(2) if Re s < σc , then n=1 an n diverges.
Note 3.7. To determine the abscissa of convergence, it is enough
to look at convergence of the series for s ∈ R.
Example 3.8. It may not be true that the series ∞ −s
P
n=1 an n con-
verges absolutely in Ωσc +δ for every δ > 0, in contrast with the behavior
of power series. An example of this phenomenon is the alternating zeta
function defined as
∞
X (−1)n
ζ̃(s) = .
n=1
ns
First note that σc = 0 for this series. Indeed, the alternating series
test implies convergence for all σ > 0, and the series clearly diverges if
P≤
σ
∞
0. Absolute convergence of the series is convergence of the series
1
n=1 nσ , so occurs if and only if <(s) > 1.
Definition 3.9. Given a Dirichlet series ∞ −s

P
n=1 an n , the abscissa
of absolute convergence is defined as
( ∞
)
X
σa = inf ρ : an n−s converges absolutely for some s with Re s = ρ
( n=1
∞
)
X
= inf ρ : an n−s converges absolutely for all s with Re s ≥ ρ .
n=1
Proposition 3.10. For any Dirichlet series, we have

σc ≤ σa ≤ σc + 1.
Proof: The first inequality is obvious. For the second, assume,
by
P∞ the usual trick, that σc = 0. We need to show that for σ > 1,
−s
n=1 |an n | converges. Take ε > 0 such that σ − ε > 1. Then,
∞ ∞ ∞ ∞
X an
X |an | X |an | 1 X 1
s = σ
= ε
· σ−ε , ≤ C σ−ε
< ∞,
n=1
n n=1
n n=1
n n n=1
n
a
where C := supn nε is finite, since σc = 0.
n

Remark 3.11. If an > 0 for all n ∈ N+ , then σc = σa . This follows

immediately by considering s ∈ R.
Recall that for the radius of convergence of a power series, we have

the following formula
1
1/R = lim sup |an | n .
n→∞
The following is an analogous formula for the abscissa of convergence

of a Dirichlet series.
Theorem 3.12. Let ∞ −s
P
n=1 an n be a Dirichlet series, and let σc
be its abscissa of convergence. Let sn = a1 + · · · + an and rn = an+1 +
an+2 + . . . .
|sn |
an diverges, then 0 ≤ σc = lim supn→∞ log
P
(1) If log n
.
|rn |
an converges, then 0 ≥ σc = lim supn→∞ log
P
(2) If log n
.
Proof: We will
Pshow (1); the second part has a similar proof. Hence
∞
we assume that n=1 an diverges and define
log |sn |
α := lim sup .
n→∞ log n
We will first prove the inequality α ≤ σc . Assume that ∞ −σ
P
n=1 an n
converges. Thus σ > 0Pand we need to show that σ ≥ α. Let
n
bn = an n−σ and Bn = k=1 bk (so that B0 = 0). By assumption,
the sequence {Bn } is bounded, say by M , and we can use summation
by parts as follows:
N
X
sN = an
n=1
N
X
= bn n σ
n=1
N
X −1
= Bn [nσ − (n + 1)σ ] + BN N σ
n=1
so that
N
X −1
|sn | ≤ M [(n + 1)σ − nσ ] + M N σ
n=1
σ
≤ 2M N .
Applying the natural logarithm to both sides yields
log |sn | ≤ σ log N + log 2M,
3. CONVERGENCE OF DIRICHLET SERIES 27
so
log |sn | log 2M
≤σ+ ,
log N log N
and this tends to σ as N → ∞, giving the desired upper bound for α.
We need to show thePother inequality: σc ≤ α. Suppose that σ > α;
we need to show that ∞ n=1 an n
−σ
converges. Choose an ε > 0 such
that α + ε < σ. By definition, there exist n0 ∈ N such that for all
n ≥ n0
log |sn |
≤ α + ε.
log n
This implies that
log |sn | ≤ (α + ε) log n = log(nα+ε ).
Thus, |sn | ≤ nα+ε , for all n ≥ n0 . Observe that
Z n+1
1 1 du
σ
− σ
= σ σ+1
≤ σn−(σ+1) .
n (n + 1) n u
Using summation by parts, we can compute
N N
X an X
sn n−σ − (n + 1)−σ + sN (N + 1)−σ − sM M −σ

=
n=M +1
nσ n=M
XN
nα+ε σn−σ−1 + N α+ε N −σ + M α+ε M −σ

≤
n=M
. (M − 1)α+ε−σ ,
and the last quantity tends to zero as M tends toR∞.
N −1
We estimated N α+ε−σ−1
by the integral M −1 xα+ε−σ−1 dx .
P
n=M n
(M −1)α+ε−σ , and the symbol . means less than or equal to a constant
times the right hand-side (where the constant depends on α + ε − σ,
but, critically, not on M ).
Exercise 3.13. Prove (2) of Theorem 3.12.

From the formulae above we can simply deduce formulae for the
abscissa of absolute convergence, although these can be derived easily
on their own.
Corollary 3.14. For a Dirichlet series ∞ −s
P
n=1 an n , we have
|+···+|an |)
|an | diverges, then σa = lim supn→∞ log(|a1log
P
(1) if n
≥ 0,
P log(|an+1 |+|an+2 |+... )
(2) if |an | converges, then σa = lim supn→∞ log n
≤
0.
Proof: Recall that to determine the abscissae, one only needs to

consider s ∈ R and then absolute convergence of the series is exactly
convergence of the Dirichlet series whose coefficient are the absolute
values of the original coefficients.
Example 3.15. The series

∞
X (−1)n
n=1
psn
has σc = 0 and σa = 1.
Proof: The series of coefficients diverges and so we use the first of
the pair of formulae for each abscissae:
log 1
σc = lim sup = 0,
n→∞ log n
and, using the prime number theorem,
log(π(n)) log n − log(log n)
σa = lim sup = lim sup = 1.
n→∞ log n n→∞ log n
Exercise 3.16.
P Show−λthat Theorem 3.1 holds for the generalized
Dirichlet series ∞ a
n=1 n e ns
, (assuming, as we always do, that λn is
an increasing sequence tending to infinity).
R (Hint:
−sx
Find a substitute for (3.3), by considering the integral
se dx .)
Therefore generalized Dirichlet series also have an abscissa of con-
vergence.
Exercise 3.17. Show that Theorem3.12 implies that if the abscissa
of convergence σc ≥ 0, then
∀ ε > 0, sn = O(nσc +ε ). (3.18)
CHAPTER 4
Perron’s and Schnee’s formulae
Suppose you know the function values f (s) of some function f that,
at
P∞ least in some half-plane, can be represented by the Dirichlet series
−s
n=1 an n . How do you determine the coefficients an ? We have seen
one way already in Proposition 1.15:
a1 = lim f (s)
s→∞
a2 = lim 2s [f (s) − a1 ]
s→∞
a3 = lim 3s [f (s) − a1 − a2 2−s ]
s→∞
and so on. The disadvantage is that these formulae are inductive.

Schnee’s theorem (Theorem 4.11) gives an integral formula for an , and
Perron’s formula (Theorem 4.5) gives a formula for the partial sums.
First, we need to recall the Mellin transform.
Definition 4.1. Suppose that g(x)xσ−1 ∈ L1 (0, ∞), then
Z ∞
(Mg)(s) := g(x)xs−1 ds
0
is the Mellin transform of g at s = σ + it.
Remark 4.2. The Mellin transform is closely related to the Fourier
transform and the Laplace transform. From one point of view, the
Fourier transform is the Gelfand transform for the group (R, +), while
the Mellin transform is the Gelfand transform for the group (R+ , ×).
The two groups are isomorphic and homeomorphic via the exponential
map, and we can use this to derive the formula for the inverse of the
Mellin transform.
Here is an inverse transform theorem for the Fourier transform.
BVloc means locally of bounded variation, i.e. every point has a neigh-
borhood on which the total variation of the function is finite.
Theorem 4.3. If h ∈ BVloc (−∞, ∞) ∩ L1 (−∞, ∞), then
Z T
1 + − 1
(Fh)(ξ)eiλξ dξ,

h(λ ) + h(λ ) = lim
2 2π T →∞ −T
29
30 4. PERRON’S AND SCHNEE’S FORMULAE
for all λ ∈ R.
Proof: See [Tit37, Thm. 24].
This gives us the following formula for the inverse of the Mellin
transform.
Theorem 4.4. Suppose that g ∈ BVloc (0, ∞). Let σ ∈ R, and
assume that g(x)xσ−1 ∈ L1 (0, ∞). Then
Z σ+iT
1 + − 1
(Mg)(s)x−s ds,

g(x ) + g(x ) = lim
2 2πi T →∞ σ−iT
for all x > 0.
Proof: Let λ = log x, then G(λ) := g(eλ ) belongs to BVloc (−∞, ∞).
Let h(λ) := G(λ)eλσ . Then
Z ∞ Z ∞
|h(λ)| dλ = |g(eλ )|eλ(σ−1) eλ dλ
−∞
Z−∞∞
= |g(x)|xσ−1 dx
0
< ∞,
so h belongs to L1 (−∞, ∞). It also belong to BVloc (−∞, ∞), be-
cause, locally, it is the product of a function of bounded variation and
a bounded increasing function. We have
Z ∞
(Mg)(s) = g(x)xs−1 dx
Z 0∞
= G(λ)eλs dλ
Z0 ∞
G(λ)eλσ eiλt dt

=
−∞
= F G(λ)eλσ (−t)

= (Fh)(−t).
Now, we apply Theorem 4.3 to h.
1 + 1
g(x ) + g(x− ) = e−λσ h(λ+ ) + h(λ− )

2 2 Z T
−λσ 1
= e lim (Fh)(ξ)eiλξ dξ
2π T →∞
Z−T
T
1
= e−λσ lim (Mg)(σ − it)eiλt dt
2π T →∞ −T
Z T
−λσ 1
= e lim (Mg)(σ + it)e−iλt dt
2π T →∞ −T
4. PERRON’S AND SCHNEE’S FORMULAE 31
Z T
1
= lim (Mg)(σ + it)e−λ(σ+it) dt
2π T →∞ −T
Z σ+iT
1
= lim (Mg)(s)e−λs ds
2πi T →∞ σ−iT
Z σ+iT
1
= lim (Mg)(s)x−s ds,
2πi T →∞ σ−iT
and we are done.
P∞ P0
Given a Dirichlet series n=1 an n−s , let F (x) = n≤x an , where
P0
means that for x = m ∈ N+ , the last term of the sum is replaced
am
by 2 so that the function F (x) satisfies
1
F (x+ ) + F (x− )

F (x) =
2
for all x. This function F (x) is called the summatory function of the
Dirichlet series.
P∞Theorem 4.5. (Perron’s formula) For a Dirichlet series f (s) =

−s
n=1 an n , the summatory function satisfies
Z σ+iT
1 f (w) w
F (x) = lim x dw, (4.6)
2πi T →∞ σ−iT w
for all σ > max(0, σc ).
Before we prove Perron’s formula, we need the following two propo-
sitions.
Proposition 4.7. Let Fσ (x) = 0n≤x an n−σ , then
P
Rx
(1) Fσ (x) = x−σ F (x) + σR 0 F (y)y −σ−1 dy,
x
(2) F (x) = xσ Fσ (x) − σ 0 Fσ (y)y σ−1 dy.
Proof: First note that if σ = 0, the formulae hold trivially.
To prove (1), evaluate the integral on the RHS by parts, assuming
that x ∈
/ N:
x Z x
−σ −σ
RHS = x F (x) + − F (y)y + y −σ dF (y)
0 0
X
−σ
= an n = Fσ (x).
n≤x
+
If x0 ∈ N , note that the difference between the limit of the LHS as
x → x0 − and the value of the LHS at x0 is 12 an n−σ and the same is
true for the RHS, since the integral on the RHS depends continuously
on x. Since the two sides were equal for all x ∈ (x0 − 1, x0 ) and they
jump by the same amount at x0 , they are equal at x0 as well.
To prove (2) one canPeither do an analogous calculation, or set

bn = an n−σ , let Gσ (x) = 0n≤x bn n−σ , and let G(x) = G0 (x). Now we
apply (1) with G in place of F and σ̃ = −σ instead of σ to get
F (x) = Gσ̃ (x)
Z x
−σ̃
= x G(x) + σ̃ G(y)y −σ̃−1 dy
Z 0x
= xσ Fσ (x) − σ Fσ (y)y σ−1 dy,
0
since F (x) = G−σ (x) and Fσ (x) = G(x).

The following is a necessary condition for a function to be repre-
sentable by a Dirichlet series.
P∞Proposition 4.8. Consider the Dirichlet series f (s) ∼

−s
n=1 an n and take a positive σ satisfying σ > σ1 := max(0, σc ).
Then
f (σ + it) = o(|t|), as |t| → ∞. (4.9)
Proof: By Theorem 3.12, we know that F (x)x−σ → 0 as x → ∞
(see Exercise 3.17). Since F (x) is 0 if x < 1, we have that F (x)x−σ−1 ∈
L1 (0, ∞). By Proposition 4.7,
Z ∞
−σ
f (σ) = lim Fσ (x) = lim x F (x) + σ F (y)y −σ−1 dy.
x→∞ x→∞ 0
Since the first term tends to 0, we obtain

f (σ)
= (MF )(−σ), for all σ > σ1 . (4.10)
σ
In fact, (4.10) holds for all s ∈ Ωσ1 , since both sides are ana-
lytic there. As The function H(λ) := F (eλ )e−λs is integrable and
F(H)(t) = (MF )(−s), by a similar change of variables argument to
the one we used in the proof of Theorem 4.4. So by the Riemann-
Lebesgue lemma we have
f (s)
lim = lim (MF )(−s)
t→±∞ s t→±∞
= lim F(H)(t)
t→±∞
= 0.
Therefore we get (4.9), since |s| ≈ |t| as t → ±∞.
We will now prove Perron’s formula (4.6).
4. PERRON’S AND SCHNEE’S FORMULAE 33
Proof: The function F is in BVloc (0, ∞), and F (x)x−σ−1 ∈

1
L (0, ∞). So we can apply the Mellin inversion formula to MF (−s) =
f (s)
s
and use the substitution u = −w as follows:
Z −σ+iT
1 − 1
+
F (x) = [F (x ) + F (x )] = lim (MF )(u)x−u du
2 2πi T →∞ −σ−iT
Z σ−iT
1 f (w) w
= − lim x dw
2πi T →∞ σ+iT w
Z σ+iT
1 f (w) w
= lim x dw,
2πi T →∞ σ−iT w
and we are done.
One can use this formula to estimate the growth of an from es-
timates of the growth of f (w). Also, note that the formula might
hold for smaller σ’s, provided that f extends holomorphically to larger
half-planes. This follows from the Cauchy integral formula applied to
integrals along long vertical rectangles.
Recall that one can use the Cauchy integral formula to obtain the
coefficients of a power series from the values of the function it repre-
sents, namely
Z 2π
1
an = f (eiθ )e−inθ dθ.
2π 0
The following theorem is a Dirichlet series analogue.
P∞Theorem 4.11. (Schnee) Consider the Dirichlet series f (s) ∼

−s
n=1 an n . One has, for σ > σc ,
Z T (
1 iλt an n−σ , if λ = log n,
lim f (σ + it)e dt = (4.12)
T →∞ 2T −T 0, otherwise.
Proof: Formally, exchanging the order of summation and integra-

tion, one gets
Z T Z T X
1 iλt
f (σ + it)e dt = − an e(−σ−it) log n+iλt dt
2T −T −T
X Z T
−σ
= an n − ei(λ−log n)t dt. (4.13)
−T
Z T
We write − to denote the normalized integral, obtained by dividing
−T
by the size of the set over which we are integrating. The integral in
(4.13) is 1 if λ = log n, and tends to 0 as T → ∞ otherwise, since for

α 6= 0, one has
Z T
1 iαT sin(αT )
− eiαt dt = e − e−iαT =

.
−T 2T iα αT
This computation works fine for finite sums, and hence we can
change finitely many coefficients of the series. So we may assume that
an = 0, if log n ≤ λ + 1, and then we must show that the LHS of (4.12)
is zero.
Case (i): σ > 0.
Consider the integral inside the limit. Then t lies in the finite inter-
val [−T, T ] and on this interval the series converges uniformly (since it
is contained in an appropriate sector, and we can apply Theorem 3.1).
Thus we may interchange the order of summation and integration and
then use integration by parts as follows
Z T X Z T
1 −σ i(λ−log n)t
X
−σ
an n e dt = an n − ei(λ−log n)t dt
2T −T −T
n≥eλ+1
Z ∞ Z T
= x−σ − ei(λ−log x)t dt dF (x)
Z0 ∞ −T
sin[(λ − log x)T ]
= x−σ dF (x)
0 (λ − log x)T
−σ ∞
x sin[(λ − log x)T ]
= F (x) (4.14)
(λ − log x)T 0
Z ∞
d x−σ sin[(λ − log x)T ]

− F (x) (4.15)
dx.
0 dx (λ − log x)T
Since F (x) = 0 for x < 1, the term in brackets in (4.14) vanishes

at 0. At infinity, F (x) = O(xσ1 +ε ) = o(xσ ), by choosing ε small
enough. (Again we let σ1 denote max(0, σc )). Hence the expression
is o((log x)−1 ) and so the whole term (4.14) vanishes.
We will show that the limit of (4.15) as T → ∞ vanishes as well.
We need to differentiate the square bracket. We obtain three terms:
d x−σ sin[(λ − log x)T ] −σx−σ−1 sin[(λ − log x)T ]

=
dx (λ − log x)T (λ − log x)T
−σ−1
−x T cos[(λ − log x)T ]
+
(λ − log x)T
x−σ−1 sin[(λ − log x)T ]
+ .
(λ − log x)2 T
4.1. NOTES 35
For each of these terms, we estimate the corresponding integral. Recall

that F (x) = O(xσ1 +ε ) for any positive ε. In particular, x−σ F (x) =
O(x−δ ), for any δ < σ − σ1 . The first and third terms are similar and
we get, as T → ∞
Z ∞
−σx−σ−1 sin[(λ − log x)T ] 1 ∞
Z
O(x−1−δ ) → 0,

I: λ+1 F (x)
dx =
Ze ∞ (λ − log x)T T
Z ∞e λ+1
−σ−1

x sin[(λ − log x)T ] 1
O(x−1−δ ) → 0.

III : λ+1 F (x)

2T
dx =
e (λ − log x) T eλ+1
The remaining term is more delicate. We will use the change of vari-
ables u = (log x − λ), so that dx = eu+λ du. We have
Z ∞ Z ∞
−x−σ−1 T cos[(λ − log x)T ] cos T u λ+u
II : F (x) dx = − F (eu+λ )e−(1+σ)(u+λ) e du
eλ+1 (λ − log x)T u
Z1 ∞
F (eu+λ )e−σ(u+λ)
= − cos T u du,
1 u
and the last integral tends to 0 as T → ∞ by the Riemann-Lebesgue
lemma. Indeed,
e−σ(u+λ)
g(u) := F (eu+λ ) = O(e−δ(u+λ) ), as u → ∞,
u
and thus belongs to L1 .
Case (ii): σ ≤ 0.
Choose some a such that σ + a > 0, and define g(s) = f (s − a).
Then
Z T Z T
1 iλt 1
f (σ + it)e dt = g((σ + a) + it)eiλt dt,
2T −T 2T −T
and we can reduce to Case (i).
Exercise 4.16. Check that the same proof yields Schnee’s theorem
for generalized Dirichlet series. Let λn be a strictly increasing sequence
with
P lim n→∞ λn = ∞. Define the abscissa of convergence for f (s) =
an e−λn s just as for an ordinary Dirichlet series. Then, for σ > σc ,
Z T (
1 iµt an e−λn σ , if µ = λn ,
lim f (σ + it)e dt = (4.17)
T →∞ 2T −T 0, otherwise.
4.1. Notes
Perron’s formula, like Schnee’s theorem, also holds for generalized
Dirichlet series. For further results in this vein, see [Hel05, Ch. 1].
CHAPTER 5
Abscissae of uniform and bounded convergence
5.1. Uniform Convergence

We introduced the alternating zeta function ζ̃ in Example 3.8, and
showed its abscissa of convergence was 0, whilst its abscissa of absolute
convergence was 1. In the strip {0 < <s < 1}, one can ask whether
there is another form of convergence, intermediate between absolute
and pointwise conditional convergence. For example, in what half-
planes does the series converge uniformly or to a bounded function?
The values of the alternating zeta function are closely related to
the values of the Riemann zeta function; more precisely,
ζ̃(s) = (21−s − 1)ζ(s). (5.1)
Indeed, for σ > 1, both series converge absolutely, so we can reorder
the terms freely, and hence
∞
X (−1)n
ζ̃(s) =
n=1
ns
∞ ∞
X −1 X 1
= +2
n=1
ns n=1
(2n)s
1−s
= (−1 + 2 )ζ(s).
We will see later [?] that ζ(s) can be analytically continued to C \ {1},
and that this continuation is unbounded on any of the lines {s : Re s =
α} with α ∈ (0, 1). The relationship (5.1) will hold for the continuation
as well, since both sides are analytic, and shows that ζ̃(s) must also be
unbounded on {Re s = α}. Hence the convergence cannot be uniform
on this line either. We can get uniform convergence, however, provided
we divide by s, as the following proposition shows.
Proposition 5.2. If f (s) = ∞ −s
P
n=1 an n converges at s0 = 0, then,
for any δ > 0,
∞
1X
an n−s
s n=1
f (s)
converges uniformly to s
in Ωδ .
37
38 5. ABSCISSAE OF UNIFORM AND BOUNDED CONVERGENCE
Proof: We use the same estimates as when we proved uniform con-

vergence in the sector, but we replace the inequality (3.4) by

|s| 1 1 |s|
ε − ≤ ε.
σ M σ (N + 1)σ δ
This is an estimate for the main term of N −s
P
n=M an n . Each of the two
other terms was estimated by ε, so with the extra 1/s, we obtain, using
1/|s| < 1/δ,
1 XN
−s ε
a n ≤ 3 ,

n
s δ

n=M
for M, N ≥ n0 and s ∈ Ωδ . Thus we are done.
Definition 5.3. For a Dirichlet series f (s) ∼ ∞ −s

P
n=1 an n we de-
fine the abscissa of uniform convergence σu as
( ∞
)
X
σu := inf ρ : an n−s converges uniformly in Ωρ ,
n=1
and the abscissa of bounded convergence σb as

( ∞
)
X
σb := inf ρ : an n−s converges to a bounded function in Ωρ .
n=1
If a Dirichlet series converges absolutely at some s0 ∈ C, then it

converges uniformly in the closed half-plane Ωσ0 by the comparison
criterion. Also, if a Dirichlet series converges uniformly in some half-
plane Ωσ0 , for N large enough, the sum differs by at most 1 from the
partial sum N −s
P
n=1 an n , for all s ∈ Ωσ0 . But (the absolute value of)
PN
this partial sum is bounded by n=1 |an |n−σ0 < ∞, and so the Dirichlet
series converges to a bounded function in Ωσ0 . Combining these two
observations with the previously known inequalities between σc and σa
and the obvious inequality σc ≤ σb we obtain
σc ≤ σb ≤ σu ≤ σa ≤ σc + 1.
In fact, σb = σu , a result due to Bohr in 1913 [Boh13b].
Theorem 5.4. (H. Bohr) Suppose that a Dirichlet series con-
verges somewhere and extends analytically to a bounded function in
Ωρ . Then for all δ > 0, the Dirichlet series converges uniformly in
Ωρ+δ .
Proof: Suppose that |f | ≤ K in Ωρ and fix 0 < δ < 1. If ρ ≥ σa ,
we are done by the chain of inequalities above. Thus, we may assume
5.1. UNIFORM CONVERGENCE 39
that ρ < σa . Observe, that it is enough to prove the following estimate

for σ ≥ ρ + δ:

N
X
f (s) − an n−s ≤ C(K, δ)N −δ log N, (5.2)

n=1
since the right-hand side is o(1) as N → ∞.

To prove 5.2, we fix s and N and define
z−s
f (z) 1
g(z) := N+ .
z−s 2
Let d denote σa − ρ + 2, and integrate g around the rectangle with
vertices s − δ ± iN d and s + (σa − ρ) ± iN d .
It would be nice to put a picture in here
By the residue theorem, we obtain
Z
g(z) dz = 2πif (s).

Consider the left-hand edge of the rectangle (LHE), on it we can esti-

mate
−δ
K 1
|g(z)| ≤ p N+
δ 2 + Im2 (z − s) 2
so that
Nd
Z Z
−δ 1
g(z) dz . KN p dy

LHE −N d δ + y2
2
p N d
−δ
= KN log y + δ 2 + y 2
−N d
−δ
≤ CKN [log N + log δ]
= C(K, δ)N −δ log N.
As for the integration over both of the horizontal edges (HE), we can
use the same estimate
Z Z σ+d−2 x−σ
−d 1
g(z) dz ≤ KN N+ dx

HE σ−δ 2
x=σ+d−2
−d 1 x−σ
. KN N
log N x=σ−δ
KN −2
. .
log N
Hence, we can conclude that

Z
2πif (s) = g(z) dz + O(N −δ log N ).
RHE
Since the series converges absolutely on RHE, we can interchange

the order of integration and summation
Z Z ∞ z−s
X
−z 1 1
g(z) dz = an n N+ dz
RHE RHE n=1 2 z−s
∞ Z z−s
X
−z 1 1
= an n N+ dz
n=1 RHE 2 z−s
∞ z−s
N + 12
Z
X
−s 1
= an n dz
n=1 RHE n z−s
We will show that the contribution of the tail of the series above
— the sum for n > N — is small, while the sum over n ≤ N is
approximately the partial sum of the Dirichlet series.
First, assume that n > N , i.e., n ≥ N + 1. Apply Cauchy’s theo-
rem to the rectangular path whose left-hand edge is RHE and whose
horizontal sides have length L, and let L tend to infinity. Since the in-
tegrand has no poles in the region encompassed by this rectangle, the
integral over the closed path vanishes. On the new right-hand edge,
the integrand decays exponentially with L and so the limit of the inte-
gral over this edge tends to 0. On the top edge (and similarly, on the
bottom one), we estimate as follows,
Z
∞+it±iN d N + 1 z−s dz 1
Z ∞
N + 12
x−σ
2
dz ≤ dx

n z−s N d σ+(σa −ρ) n

s+(σa −ρ)±iN d
N+ 1
Z ∞
1 (x−σ) log n
2
= e dx
N d σ+(σa −ρ)
N+ 1

1 1 (σa −ρ) log n
2
= e .
N d − log N + 12

n
N + 12

The expression log n
is minimized when n = N + 1. So
1
N+ 2 1
| log | ≥ − log(1 − )
n 2N + 2
1
> .
2(N + 1)
5.1. UNIFORM CONVERGENCE 41
Hence, we can estimate the tail of the series by

∞ 1 z−s

∞ σa −ρ
|an | 1−d N + 21

N +
Z
X
−s 2 1 X
a n dz . N

n σ
n z−s n n

n=N +1 RHE n=N +1
σa −ρ X∞
1−d 1 |an |
= N N+
2 n=N +1
n a −ρ
σ+σ
. N −1 ,
P |an |
since nσ+σa −ρ
converges.
If n ≤ N , we use Cauchy’s theorem again, but now with a rectan-
gular path whose right-hand edge is RHE and whose width L tends to
infinity. The residue theorem now implies that
z−s
N + 12
Z
1
dz = 2πi.
n z−s
The integrand decays exponentially on the left-hand edge, and so the
integral over that edge tends to zero. As for the top edge (and also the
bottom one)
Z
s+(σa −ρ)±iN d N + 1 z−s dz 1
Z σ+(σa −ρ)
N + 12
x−σ
2
dz ≤ dx

n z−s N d −∞ n

−∞+it±iN d
N+ 1
Z σ+(σa −ρ)
1 (x−σ) log n
2
= e dx
N d −∞
N+ 1

1 1 (σa −ρ) log n
2
= e
N d log N + 12

n
σ −ρ
N + 12 a

−d 1
≤ N 1
N+
log N 2 n
σ −ρ
N + 12 a

1−d
. N
n
. N −1 n−σa +ρ
Thus,
N 1 z−s N N
N+ |an |
Z
X
−s 2 1 X an −1 −σa +ρ
X
an n dz = 2πi + O(N n )
n=1 RHE n z−s n=1
ns n=1
nσ
N
X an
= 2πi + O(N −1 ),
n=1
ns
where we used boundedness of the partial

R sums of the convergent series
P |an | 1
n nσ+σa −ρ . We have shown that 2πi RHE g(z) dz is close to both the
partial sum of the Dirichlet series and f (s) (and the error is as in (5.2),
and does not depend on s).
The promised equality of the two new abscissae is now an immediate
corollary.
Corollary 5.5. The equality σb = σu holds for any Dirichlet se-
ries.
Note, however, that the above corollary does not imply that if a
Dirichlet series converges to a bounded function in some half-plane, it
will converge uniformly in that half-plane. We only know that it will
converge uniformly in every strictly smaller half-plane.
Remark 5.6. The function g(z) used in the proof of the theorem
above comes from Perron’s formula which can be restated as (in the
special case of x = N + 21 )
(N + 12 )z−s
Z σ+it+iT
X
−s 1
an n = f (z) dz + eN,T ,
n≤N
2πi σ+it−iT z − s
where eN,T is an error term that comes from not taking the limit in T .
One can also prove this formula using the estimates above.
5.2. The Bohr correspondence

Bohr’s idea was to use the following correspondence between Dirich-
let series and power series in infinitely many variables. For a positive
integer with prime factorization n = pk11 . . . pkl l , we define
z r(n) := z1k1 . . . zlkl .
We have an isomorphism between formal power series in infinitely
many variables z1 , z2 , . . . and Dirichlet series, given by
X X
B: an z r(n) 7→ an n−s . (5.7)
n n
We shall write Q for the inverse of B:

X X
Q: an n−s 7→ an z r(n) . (5.8)
n n
The map B is an evaluation homomorphism — indeed, we evaluate

the power series on the one-dimensional set {(zi ) : zi = p−s i }. It is
clearly onto, and it has a trivial kernel because the right-hand side is
0 iff all the coefficients vanish.
5.3. BOHNENBLUST-HILLE THEOREM 43
For finite series, we can norm both spaces so that B will be isomet-
ric. Indeed, we have
N N
Z T
X 2
1 −it
X
lim an n dt = |an |2
T →∞ 2T −T

n=1 n=1
Z N 2
X 2πit·r(n)

=
a n e dt. (5.9)

T ∞
n=1
∞
By T we mean the infinite torus
T∞ = {(e2πit1 , e2πit2 , . . . ) : 0 ≤ tj < 1 ∀j ∈ N+ }
which we identify with the infinite product
[0, 1) × [0, 1) × · · ·
on which we put the product probability measure of Lebesgue measure
on each interval.
We shall investigate when (5.9) holds for infinite sums in Theorem
6.39.
Flesh this section out.
5.3. Bohnenblust-Hille Theorem

We will now proceed to show that σa − σb ≤ 21 , and that this bound
is sharp. Originally, Hille and Bohnenblust exhibited an example of a
Dirichlet series for which equality holds in the above inequality. Their
construction was extremely complicated.
Instead of going through their construction, we shall show that
such an example exists using a probabilistic method. This is a non-
constructive method, used in other fields, in particular in combina-
torics/graph theory.
Before describing the probabilistic method we mention two anal-
ogous methods: the “cardinality method” and the “Baire category
method”. Recall that one can prove the existence of transcendental
numbers by showing that there are only countably many algebraic num-
bers (and uncountably many real numbers). This is much easier than
proving that a concrete number is transcendental. Similarly, the ex-
istence of a nowhere differentiable continuous function on an interval
I can be proved by showing that the set of all continuous functions
with a derivative at at least one point is of the first category (and thus
cannot equal the complete metric space of all continuous function on
I). The construction of a particular example is again fairly technical.
The probabilistic method is similar in spirit. Instead of exhibiting

a concrete example of an object with some given property, we con-
sider some set S of objects and equip it with a convenient probability
measure. We strive to show that a randomly chosen object will have
the desired property with a non-zero probability. Although it might
seem that this will rarely work, the probability method has been very
successful, especially when examples with the given property have a
complicated structure or description.
In our case, we will need to consider random series of functions of
the form
X∞
fε (s) = εn an n−s ,
n=1
where {εn } is a Rademacher sequence, that is, a sequence of indepen-
dent random variables, such that each εn ∈ {±1} and Prob (εn = 1) =
Prob (εn = −1) = 1/2. One can also consider the random series
∞
X
fω (s) = an einωn n−s ,
n=1
where ω = {ωn }∞ n=1 is a sequence of random variables that are in-

dependent and such that each ωn is uniformly distributed on [0, 2π].
Both {εn } and {ωn } are i.i.d.’s, that is, independent and identically
distributed.
When we have a Rademacher sequence, we use E to denote the
expectation, that is the average over all choices of sign, of some function
that depends on the sequence:
X
E[ εn gn ].
If the sequence is finite of length K, this just means adding up all
2K choices and dividing by 2K . If the sequence is infinite, one must
replace this by integrating over the space {−1, 1}∞ with the product
probability measure.
Note that a sequence of i.i.d.’s has a canonical probability distribu-
tion associated to it, namely the product probability. Heuristically, to
choose a random sequence, we can choose it element by element, and
since these elements should be independent, we arrive at the product
probability.
As an example of a theorem about random Dirichlet series we prove
the following proposition.
Proposition 5.10. Let {an } be a sequence of complex numbers and
let {ωn }∞
n=1 be sequence of i.i.d.’s which are uniformly distributed on
[0; 2π]. Denote fω (s) := ∞ inωn −s

P
n=1 an e n , as above. Then there exists
some σ̃ = σ̃({an }) such that σc (fω ) = σ̃ almost surely.
Proof: Given a sequence of random variables, a tail event is an
event whose incidence is not changed by changing the values assumed
by any finitely many elements of the sequence. The zero-one law of
probability asserts that any tail event associated to a sequence of i.i.d.’s
happens with probability either 0 or 1 [Kah85, p.7]. Consider the
events Ba = {fω : σc (fω ) ≤ a} for a ∈ R. These are clearly tail events.
Let
σ̃ := inf {a : Prob (Ba ) = 1},
where we agree that inf ∅ = ∞. Since the events Ba are nested, we
have
Prob (Ba ) = 0, for all a < σ̃,

Prob (Ba ) = 1, for all a > σ̃,
T S
and {fω : σc (fω ) = σ̃} = n Bσ̃+ 1 \ n Bσ̃− 1 ,
n n
which easily implies that σ̃ has the desired property.
Let f be a holomorphic function on a domain Ω ⊂ C. We say that

∂Ω is a natural boundary for f , if no point z0 ∈ ∂Ω has a neighborhood
to which it can be holomorphically continued. Proposition 5.10 can be
strengthened in the following way [Kah85, p. 44].
Theorem 5.11. Let ωn and σ be as above. Then, with probability
1, the line {Re s = σ} is the natural boundary for the Dirichlet series
iωn −s
P
an e n .
We will need the following theorem, which we shall prove as Corol-
lary 5.23 below. We shall use multi-index notation, where α ∈ Zr —
see Appendix 11.1. We shall use Tr to denote the r-torus, which by
an abuse of notation we shall identify with both {(e2πit1 , · · · , e2πitr ) :
0 ≤ tj ≤ 1 ∀ j} and {(t1 , · · · , tr ) : 0 ≤ tj ≤ 1 ∀ j}.
Theorem 5.12. There exists a universal constant C > 0 such that
for every r ∈ N+ , every N ≥ 2, and every choice of coefficients cα ∈ C,
with |α| = |α1 | + · · · + |αr | ≤ N , there exists some choice of signs such
that

X
h X i1
2 2
2πi(α t +···+α t )

sup ±cα e 1 1 r r
≤ C r log N |cα | . (5.13)
t∈Tr
|α|≤N
By Fubini’s theorem and the orthogonality of {e2πiα·t }, for any

choice of signs
Z X 2
X
2πiα·t

±c n e dt = |cα |2 ,
r T α α
so the left-hand side of (5.13) is at least α |cα |2 . The theorem says

P
that
√ for some choice of signs, this estimate is only off by a factor of
r log N .
Note that choosing all cα positive and using the Cauchy-Schwarz
inequality yields the following much cruder estimate:

X
i(α1 t1 +···+αr tr )
X
sup cα e = cα
t∈Tr

|α|≤N α
1
p X 2
2
≤ CN |cα | ,
α
where CN is the number of terms, roughly N r , if N r.

We will need the following lemma.
Lemma 5.14. Let
X
P (t) = cα e2πi(α1 t1 +···+αr tr )
|α|≤N
be a trigonometric polynomial on Tr . If P is real, then there ex-

ists an r-dimensional cube I ⊂ Tr of volume (N + 1)−2r on which
|P (t1 , . . . , tr )| ≥ 21 kP k∞ .
Proof: By multiplying P by (−1), if necessary, we may assume that
there exists θ = (θ1 , . . . , θr ) ∈ Tr such that
P (θ) = kP k∞ .
By the mean value theorem, we conclude that for any t = (t1 , . . . , tr ) ∈
Tr , there exists θ̃ belonging to the segment connecting t and θ such
that
r
X ∂P
P (t) − P (θ) = (tj − θj ) (θ̃),
j=1
∂tj
Thus,
r
X ∂P
|P (t) − P (θ)| ≤ max |tj − θj | ∂tj (θ̃)
(5.15)
j
j=1
There exists a choice of signs sj ∈ {±1} so that

d X ∂P
P (θ̃1 + s1 x, . . . , θ̃r + sr x) = ∂tj (θ̃) .

dx |x=0 j
We fix this choice, and define a trigonometric polynomial of degree at

most N
Q(x) = P (θ̃1 + s1 x, . . . , θ̃r + sr x).
Then Q(x) = k bk eikx and Q0 (x) = k ikbk eikx . Note that by inte-
P P
grating against e−ikx we obtain |bk | ≤ kQk∞ , and hence
X
|Q0 (0)| ≤ |kbk |
k
N
X
≤ max |bk | k
k
k=−N
≤ kQk∞ N (N + 1)
≤ kP k∞ N (N + 1).
Thus, we can continue our estimate from (5.15)
|P (t) − P (θ)| ≤ kP k∞ N (N + 1) sup |tj − θj |. (5.16)
j
Since |P (θ)| = kP k∞ , whenever the right-hand side of (5.16) is

bounded by 12 kP k∞ , we have P (t) ≥ kP2k∞ . This will occur if
1
sup |tj − θj | ≤ .
j 2N (N + 1)
The set of such t’s is a cube of volume [N (N + 1)]−r ≥ (N + 1)−2r .
Theorem 5.17. Let {Pn }K n=1 be a finite set of complex trigono-

metric polynomials in r variables of P degree less than or equal to N ,
with N ≥ 1. Let Q(t1 , . . . , tr ) = n εn Pn (t1 , . . . , tr ), where εn is a
Rademacher sequence. Then
21 !
X 2
Prob kQk∞ ≥ 32r log γN kPn k2∞ ≤ ,
n
γ
for all real γ ≥ 8.

Proof: First suppose that all Pn ’s are real, let τ = n kPn k2∞ and
P
M = kQk∞ (here M = M (ε) is a random variable). Let λ be an
x2
arbitrary real number. Then, using the inequality 12 (ex + e−x ) ≤ e 2
yields
P
E eλQ(t) = E eλ n εn Pn (t)

!
Y
= E eλεn Pn (t)
n
Y
E eλεn Pn (t)

=
n
Y 1
λPn (t) −λPn (t)
= [e +e ]
n
2
2
2 Pn (t)
Y
≤ eλ 2
n
λ2
kPn k2∞
Y
≤ e 2
n
λ2
kPn k2∞
P
= e 2 n
τ λ2
= e 2 . (5.18)
By Lemma 5.14, there exists an interval I = I(ε) ⊂ Tr of volume at
least (N + 1)−2r such that |Q| ≥ 12 kQk∞ of I. For fixed ε = {εn } we
thus have
Z
λM (ε) 1
e 2 ≤ eλQ(t) + e−λQ(t) dt
vol(I(ε)) ZI(ε)
≤ (N + 1)2r eλQ(t) + e−λQ(t) dt
Tr
Taking the expected value and using estimate (5.18) yields

λM Z
2r λQ(t) −λQ(t)
E e 2 ≤ (N + 1) E e +e dt
Z Tr
= (N + 1)2r E eλQ(t) + e−λQ(t) dt

ZTr
τ λ2
≤ (N + 1)2r 2e 2 dt
Tr
τ λ2
2r
= 2(N + 1) e 2
τ λ2
+log 2+2r log(N +1)
= e 2
Thus,
λM λ2 τ
E e 2 − 2 −log 2−2r log(N +1) ≤ 1,
and hence, by Chebyshev’s inequality,

λM λ2 τ 1
Prob e 2 − 2 −log 2−2r log(N +1) ≥ γ ≤ . (5.19)
γ
The event on the left-hand side of (5.19) is equivalent to
λM − λ2 τ
− log 2 − 2r log(N + 1) ≥ log γ. (5.20)
2
q
Choose λ = τ2 log[2γ(N + 1)2r ], then, after algebraic manipulations,
(5.20) becomes
r
2
M log[2γ(N + 1)2r ] ≥ 4 log[2γ(N + 1)2r ],
τ
which is the same as
√ p
M ≥ 2 2τ log[2γ(N + 1)2r ]. (5.21)
For γ ≥ 8 we have
2γ(N + 1)2r ≤ (γN )2r ,
so (5.21) will hold if
√ p
M 2 2τ log[γN ]2r
≥
p
= 4 rτ log[γN ].
Recalling that M = kQk∞ and τ = n kPn k2 we obtain
P
21 !
X 1
Prob kQk∞ ≥ 4 r log[γN ] kPn k2 ≤ ,
n
γ
when Q is real.
If Q is complex and
X 1
kPn k2∞ 2 ,

kQk∞ ≥ 4 2r log[γN ]
n
then one of the two following inequalities must hold:
X 21
2
kRe Qk∞ ≥ 4 r log[γN ] kRe Pn k∞ ,
n
X 21
2
kIm Qk∞ ≥ 4 r log[γN ] kIm Pn k∞ .
n
But since these inequalities involve real polynomials, either of them
happens with probability at most γ1 , by the real case. The probability
that at least one of them happens is thus at most γ2 .
Corollary 5.22. Let N ≥ 2, and let cα ∈ C be given for every

α = (α1 , . . . , αr ) ∈ Zr with |α| ≤ N . Then, for any γ ≥ 8, there exists
C > 0 such that
 
X X 21
1 2
Prob  εα cα eiα·t
≥ C(r log N )
2 |cα |2  ≤ .

∞ γ
|α|≤N |α|≤N

log γ
Proof: Fix γ ≥ 8, and choose C > 0 such that C 2 ≥ 32 1 + log N
.
iα·t
Let Pα (t) = cα e and use Theorem 5.17.
Corollary 5.23. There exist a choice of signs {εα } such that

X X 21
1
iα·t 2
ε α cα e ≤ C(r log N ) |cα | .
2

|α|≤N ∞ |α|≤N
Proof: For any γ ≥ 8, the probability that a random series will not
have the property is at most γ2 < 1.
Theorem 5.24. (H. Bohr) For any Dirichlet series σa − σu ≤ 21 .
Proof: Let ρ > σu , then ∞ −s

P
n=1 an n converges uniformly in Ωρ .
1
Fix s ∈ C with Re s = ρ + 2 + ε. By the Cauchy-Schwarz inequality,
1
X X
|an n−s | = |an |n−(ρ+ 2 +ε)
n n
X 12 X 21
−2ρ −(1+2ε)
≤ |an | n , (5.25)
n n
where the second sum converges. By uniform convergence, there exists

K > 0 such that for every t ∈ R and N ∈ N+
N
X
an n−(ρ+it) ≤ K.

n=1
Consequently,
N 2
2
X −(ρ+it)

K ≥
a n n

n=1
N
m
X X
= |an |2 n−2ρ + 2 Re an am (nm)−ρ eit log n .
n=1 1≤n<m≤N
Taking the normalized integral yields

N Z T
m
X X
2 2 −2ρ −ρ
K ≥ |an | n + 2 Re an am (nm) − eit log n dt.
n=1 1≤n<m≤N −T
Taking the limit as T tends to ∞, the mixed terms tend to 0 and so

we conclude that
XN
|an |2 n−2ρ ≤ K 2 ,
n=1
for all N ∈ N+ . Thus
P the −s first sum on the right-hand side of (5.25) is
bounded, and so n |an n | converges. Thus, σa ≤ 21 + ρ + ε. Since
this is true for every ρ > σu and ε > 0, we get σa ≤ 21 + σu .
Theorem P 5.26. (Bohnenblust-Hille, 1931) There exist a

Dirichlet series ∞ n=1 an n
−s
for which σu = 21 and σa = 1.
We shall present a probabilistic proof, due to H. Boas [Boa97].
Proof: Each an will be an element of {±1, 0} and the coefficients
will be constructed in groups, starting with k = 2. To construct the k th
group, choose a homogeneous polynomial Qk of degree k in 2k variables
with coefficients εj ∈ {±1}, with j = (j1 , . . . , j2k ),
j
X
Qk (z1 , z2 , . . . , z2k ) = εj z1j1 . . . z22kk
|j|=k
so that
X 12
k 2
kQk k∞ ≤ C 2 log k |εj | .
|j|=k
This is possible, by Corollary 5.23. By Lemma 5.29, the number of

k
(monic) monomials of degree k in 2k variables is 2 +k−1 k
. We conclude
that
k 1
k 2 +k−1 2
kQk k∞ ≤ C 2 log k .
k
We convert the Qk ’s into Dirichlet series as in (5.7)
X j −s
j
fk (s) := (BQk )(s) = εj p21k . . . p22kk+2k −1 ,
|j|=k
P∞
and let f = k=2 fk , thought of as a Dirichlet series. Then the coef-
ficients of f lie in {±1, 0}, since each n can appear in at most one fk .
Claim 1: σa (f ) = 1.
Proof: In fk , the number of non-zero coefficients is

k 2
(2k )k 2k

2 +k−1
≥ ≥ k.
k k! k
By the prime number theorem, pk ≈ k log k, so that p2k+1 ≤ M 2k k, for
some M > 1. Hence any n that has a non-zero coefficient in fk must
satisfy
k
n ≤ M 2k k .
Thus, we can estimate for σ < 1,
X X 2k2 −kσ
−σ
|an |n ≥ k
M 2k k (5.27)
n k
k
2 (1−σ)
X 2k
=
k
k k(1+σ) M kσ
By the root test (or ratio test), (5.27) diverges for σ < 1.
P Since
−σ
for σ > 1 the series converges absolutely (by comparison to
nn ), we conclude that σa = 1.
Claim 2: σu (f ) = 12 .
Proof: Fix ε > 0, let σ = 21 + ε, and note that
|fk (σ + it)| = |Qk (p−s
2k
, . . . , p−s
2k+1 −1
)|

X j1 j 2k σ j1 j 2k it
=
εj p2k . . . p2k+1 −1 p2k . . . p2k+1 −1 .
|j|=k
Thus,
sup |fk (σ + it)| ≤ sup |Qk (z1 , . . . , z2k )|
t |zi |=p−σ
k , i=1,...,2k
2 −1+i
≤ p−kσ
2k
sup |Qk |
|zi |=1
1
2k + k − 1 2

≤ Cp−kσ
2k
2 log kk
(5.28)
k
h i1
k −kσ k k2 2
. (2 k log 2) 2 log k2
2 1 1 p
= (k log 2)−kσ 2k (−σ+ 2 + 2k ) log k
2 1 p
= (k log 2)−kσ 2k (−ε+ 2k ) log k.
P
The
P series k |fk | is thus estimated by a summable series. Hence,
k fk converges to a holomorphic function which is bounded in Ω1/2+ε
5.4. NOTES 53
and equal to f in Ω1 . Letting ε → 0+ yields, by Theorem 5.4, σu =

σb ≤ 12 . Thus, by Theorem 5.24 and Claim 1, σu = 12 .
Lemma
n+m−1
5.29. The number of monomials of degree m in n variables
is m
.
Proof: From a linear array of n + m − 1 objects, choose n − 1 and
color them black. Let the power of zi be the number of non-colored
objects between the (i − 1)st black one and the ith one. 2
Exercise 5.30. Fill in the details that the series in (5.28) con-
verges.
Exercise 5.31. Show that for all x ∈ [0, 12 ], there is a Dirichlet
series such that σa − σu is exactly x.
(Hint: Although Bohnenblust and Hille did not spot it, this result
is a one-line consequence of Theorem 5.26. If you find the right line!)
5.4. Notes
The proofs of the Bohnenblust-Hille theorem in Section 5.3 and
Bohr’s Theorem 5.4 are based on H. Boas’s article [Boa97]. The orig-
inal proofs are in [BH31] and [Boh13b], respectively. Theorem 5.24
was proved in [Boh13a].
Talk about recent advances, in particular [DFOC+ 11].
CHAPTER 6
Hilbert Spaces of Dirichlet Series
6.1. Beurling’s problem: The statement

We will motivate our discussion by√considering a problem posed by
A. Beurling in 1945. If we set β(x) = 2 sin(πx), the set
{β(nx) : n ∈ N+ }
forms an orthonormal basis of L2 ([0; 1]).
Proposition 6.1. If ψ : R+ → C is 2-periodic, and {ψ(nx)}n∈N+

is an orthonormal basis for L2 ([0; 1]), then ψ = eiθ β, for some θ ∈ R.
Proof: Extend ψ to an odd function on R. Then P ψ is odd and

2-periodic, so we can expand it into a sine series ψ(x) = ∞
k=1 ck β(kx).
Since {ψ(nx)}n∈N+ is an orthonormal basis, we have
∞
X
1 = kβ(mx)k2 = |hβ(mx), ψ(nx)i|2
n=1
∞
X ∞
X
= |hβ(mx), ck β(nkx)i|2
n=1
X∞ X k=12

=
ck
n=1 k; kn=m
X
= |ck |2 .
k|m
Letting m = 1, we obtain |c1 |2 = 1. Thus, for m ≥ 2, we have 1 + · · · +

|cm |2 = 1 (where the middle terms are non-negative) and so |cm | = 0.
Definition 6.2. Let {vn } be a set of vectors in a Hilbert space

H. We say that {vn } is a Riesz basis, if span {vn } = H and the Gram
matrix G given by
Gij := hvj , vi i
55
56 6. HILBERT SPACES OF DIRICHLET SERIES
is bounded and bounded below, that is, for all {an }∞ 2

n=1 ∈ ` :
∞
X ∞
X X∞
2
c1 |aj | ≤ ai aj Gij ≤ c2 |aj |2 . (6.3)
j=1 i,j=1 j=1
Proposition 6.4. The set {vn }∞

n=1 is a Riesz basis if and only if
the map
∞
X X∞
T : an en 7→ an vn
n=1 n=1
is bounded and invertible, where {en } is an orthonormal basis for H.
Proof: We have
2 2
X X X
T a e
n n
= a v
n n
= an am Gmn

n n m,n
and 2
X X

a e
n n
= |an |2 .
n n
Thus condition (6.3) is equivalent to boundedness of T from below and
above. Moreover, T is onto if and only if the span of {vn } is dense in
H. The claim follows by recalling that a map is invertible if and only
if it is bounded, bounded from below, and onto.
Here is Beurling’s question.
Question 6.5. (Beurling) For which odd 2-periodic functions ψ :
R → C does the sequence {ψ(nx)}∞ 2
n=1 form a Riesz basis for L ([0; 1])?
Remark 6.6. A frame is a set of vectors {vn }∞ n=1 in H such that

for some c1 , c2 > 0
∞
X
2
c1 kvk ≤ |hv, vn i|2 ≤ c2 kvk2
n=1
holds for every v ∈ H. (Unlike a Riesz basis, they do not need to be

linearly independent).
The following problem attracted a lot of attention; it has many
equivalent reformulations.
Conjecture 6.7. (Feichtinger) Suppose that {vn }∞ n=1 is a set of
unit vectors in H that form a frame. Does it follow that {vn }∞
n=1 is a
finite union of Riesz bases?
The conjecture was proved, in the affirmative, by A. Marcus, D.
Spielman and N. Srivastava [MSS15].
6.1. BEURLING’S PROBLEM: THE STATEMENT 57
Beurling’s idea was to consider the Hilbert space of Dirichlet series

X∞ X
2 −s 2
H := an n : |an | < ∞ . (6.8)
n=1 n
Let us first observe that for any f ∈ H2 we have σa ≤ 21 . Indeed, by

the Cauchy-Schwarz inequality,
X 21 X 12
X −s
2 −2σ

an n ≤ |an | n < ∞,
n n n
whenever 2σ > 1. In fact, the above estimate shows that for any
s0 ∈ Ω1/2 , the map H2 3 f 7→ f (s0 ) is a bounded linear functional.
Therefore it is given by the inner product with a function ks0 ∈ H2 ,
the so-called reproducing kernel at s0 , i.e.,
f (s0 ) = hf, ks0 i for all f ∈ H2 .
For any Hilbert (or Banach) space of analytic functions X , we define
its multiplier algebra by

Mult (X ) = ϕ; ϕf ∈ X , ∀f ∈ X .
It is easy to check that the following hold
• 1 ∈ X =⇒ Mult (X ) ⊂ X ,
• Mult (X ) is an algebra.
Clearly, multiplication by k −s is isometric on H2 , for all k ∈ N. Con-
sequently, every finite Dirichlet series lies in Mult (H2 ).
1
Also, note that sups∈Ω1/2 |k −s | = k − 2 → 0 as k tends to ∞. Thus,
kf kMult (H2 ) 6. kf kH ∞ (Ω1/2 ) .
The following result — multiplication operators are bounded if they
are everywhere defined — is true in great generality (see Section 11.4).
Proposition 6.9. The multiplication operator Mϕ is bounded on
H2 for every ϕ ∈ Mult (H2 ).
Proof: Multiplication operators on a Banach space of functions in
which norm convergence implies pointwise convergence (or at least a.e.
convergence) are easily seen to be closed. Indeed, suppose that fn → f
and Mϕ fn → g. Then, for every s ∈ Ω1/2 , fn (s) → f (s) and so
(Mϕ fn )(s) = ϕ(s)fn (s) → ϕ(s)f (s) = (Mϕ f )(s). On the other hand,
Mϕ fn (s) → g(s), for all s ∈ Ω1/2 . We conclude that (Mϕ f )(s) = g(s)
for all s ∈ Ω1/2 and hence Mϕ f = g. Thus, Mϕ is closed. Hence,
Mϕ is an everywhere defined closed linear operator on a Banach space,
and the closed graph theorem states that such operators are necessarily
bounded.
Now let ψ be an odd 2-periodic P function on R. We can ex-

∞
pand it into a Fourier series ψ(x) = n=1 cn β(nx). The sequence
2
{ψ(kx)}k∈N
P+ is a Riesz basis,
P if and only if it spans L and the oper-
ator T : k ak β(kx) 7→ k ak ψ(kx) is bounded and bounded below.
Denote ψk (x) := ψ(kx) and analyze the condition on T :
2 2
X X X

a k ψk

=
a k c n β(nkx)

k X k n
X
= ak cn β(nkx), aj cm β(mjx)
k,n j,m
X
= ak c n aj c m ,
k,n,j,m; kn=jm
and thus we want

X X
ak c n aj c m ≈ |ak |2 . (6.9)
k,n,j,m; kn=jm k
Let us define auxilliary functions in H2 :

X X
g(s) := cn n−s , f (s) := ak k −s .
n k
We have
X X X
−s −s
kgf k2H2 = cn ak (nk) , cm aj (mj) = ak c n aj c m ,
n,k m,j k,n,j,m; kn=jm
and so the condition (6.9) holds, if and only if kgf kH2 ≈ kf kH2 , i.e.,
when Mg is bounded and bounded below.
Let us also look the density of the span of {ψn }n . It is equivalent
to
X X
2
span cn β(nkx) = L ([0; 1]) ⇐⇒ span cn enk = `2 (N)
n k∈N+ n k∈N+
X
−s
⇐⇒ span cn (nk) = H2
k∈N+
n
⇐⇒ span k −s g(s) = H2 .
k∈N+
The last condition implies that range of Mg is dense. But since Mg is

bounded below, it has a closed range and thus is onto. Therefore Mg
is invertible, or M1/g is bounded. Conversely, if Mg is invertible, the
image of the dense set span {k −s }k∈N+ is dense, and so the density of
span {ψk }k follows by the above equivalences. We have proved:
6.2. RECIPROCALS OF DIRICHLET SERIES 59
Proposition 6.10. Let ψ(x) = ∞

P
n=1 cn β(nx) be a odd 2-periodic
function on R. Then {ψ(kx)}k∈N+ is a Riesz basis,
P∞ if and only if both
2 −s
g and 1/g are multipliers of H , where g(s) = n=1 cn n .
In view of Proposition 6.10, Beurling’s question 6.5 would be an-
swered if we could answer the following question:
Question 6.11. What are the multipliers of H2 ?
6.2. Reciprocals of Dirichlet Series

Proposition 6.12. If f (s) = ∞ −s
P
n=1 an n is a Dirichlet series that
1
converges somewhere and satisfies a1 6= 0, then g(s) = f (s) is also
given by the sum of a somewhere-convergent
Dirichlet series. Moreover,
σb (g) = inf{ρ : inf |f |Ωρ > 0}.
Proof: By rescaling, we may assume that a1 = 1, and by shifting

the series so that σa < 0, we have sup |an | ≤ M . We will construct the
coefficients bk of g inductively. Clearly, b1 = 1. For n ≥ 2, we have
X
0 = fcg(n) = an/k bk . (6.13)
k|n
Equations (6.13) can be solved for bk , first when k is a prime, then a

power of a prime, then when k has two distinct prime factors, and so
on.
Claim: If n = pi11 . . . pirr , then |bn | ≤ n2 M |i| .
Proof: For n = 1, bn = 1 and so the claim holds. Assume induc-
tively that the claim holds for all m < n. By (6.13), we have
X
|bn | ≤ |ak bn/k |
k|n, k≥2
X
≤ M bn/k
k≥2
X n 2
≤ M M |i|−1
k≥2
k
X 1
≤ M |i| n2
k≥2
k2
2
|i| 2 π
= M n −1 ,
6
π2
and the claim follows, since 6
< 2.
Since |i| ≤ log2 n, we obtain

|bn | ≤ M |i| n2
≤ M log2 n n2
= nlog2 M n2
= n2+log2 M .
Hence, for Re s > 3 + log2 M , the Dirichlet series n bn n−s converges
P
absolutely.
Now, g is bounded in Ωρ , if and only if inf |f |Ωρ > 0. As g is given
by a convergent Dirichlet series in Ω3+log2 M , by Theorem 5.4,

σb (g) ≤ inf{ρ : inf |f | > 0}. Ωρ
The reverse inequality is obvious.

Note that the condition a1 6= 0 is necessary, since a1 = limσ→∞ f (σ).
6.3. Kronecker’s Theorem

Theorem 6.14. (Kronecker)
(1) Let θ1 , . . . , θk ∈ R be linearly independent over Q, and let
α1 , . . . , αk ∈ R, T, ε > 0 be given. Then there exist t > T
and q1 , . . . , qk ∈ Z such that
|tθj − αj − qj | < ε, 1 ≤ j ≤ k.
(2) Let 1, θ1 , . . . , θk ∈ R be linearly independent over Q, and let
α1 , . . . , αk ∈ R, T, ε > 0 be given. Then there exist N 3 n > T
and q1 , . . . , qk ∈ Z such that
|nθj − αj − qj | < ε, 1 ≤ j ≤ k.
Proof: (1) =⇒ (2): Assume that all θj ’s lie in (−M, M ). Fix 0 <
ε < 1, and apply (1) to the (k + 1)-tuples θ1 , . . . , θk , 1 and α1 , . . . , αk , 0,
T = N + 1 and ε/(M + 1). Let n = qk+1 , then |t − n| < ε/(M + 1).
Thus, for 1 ≤ j ≤ k, we have
|nθj − αj − qj | ≤ |n − t|θj + |tθj − αj − qj |
Mε ε
< + .
M +1 M +1
To prove (1), define F (t) := 1 + kj=1 e2πi[θj t−αj ] . We need to
P
show that lim supt→∞ |F (t)| = k + 1. Fix m ∈ N, and define

α = (0, α1 , . . . , αk ), θ = (0, θ1 , . . . , θk ) and j = (j0 , . . . , jk ). Then
X
[F (t)]m = aj e2πitγj ,
|j|=j0 +···+jk =m
6.4. POWER SERIES IN INFINITELY MANY VARIABLES 61
where a = m! e−2πij·α and γj = j · θ. Indeed, there are m! ways to get

Q 2πitjjθ j! j!
le in the product, and,Pby independence of θj ’s over Q, distinct
l l
m
j’s yield distinct γj ’s. Also, |j|=m |aj | = (k + 1) , since there are
(k + 1) terms, each with a coefficient of modulus 1.
Suppose that lim supt→∞ F (t) < k + 1. Then there exist M > 0
and λ < k + 1 such that |F (t)| ≤ λ for all t > M . Consequently,
1 T
Z
lim sup |F (t)|m dt ≤ λm .
T →∞ T 0
Since [F (t)]m is a finite combination of exponentials,

1 T
Z
m −2πitγj

|aj | = lim [F (t)] e dt
T →∞ T 0
1 T
Z
≤ lim sup |F (t)|m dt
T →∞ T 0
≤ λm . (6.15)
Note that there are m+k

k
≤ (m + 1)k possible j’s. Thus, summing the
inequality (6.15) over all j’s yields
X
(k + 1)m = |aj |
|j|=m
≤ (m + 1)k λm ,
a contradiction for large m.
Remark 6.16. Let q1 , . . . , qk be distinct primes. Then

log q1 , . . . , log qk are linearly independent over Q.
P Proof: If not, then for some rational numbers r1 , . . . , rk we have

j rj log qk = 0 and by clearing the denominators, there exists integers
n1 , . . . , nk so that
X Y n
nk log qk = 0 =⇒ qk k = 1.
k k
Thus all nk ’s must be zero by the uniqueness of prime factorization.
6.4. Power series in infinitely many variables

Recall from (5.8) that given f ∈ H2 , f = ∞ −s
P
n=1 an n , we have a
formal power series in infinitely many variables
X∞
(Qf )(z) = an z r(n) .
n=1
Let D∞ denote {(zi )∞

i=1 ; |zi | < 1} — the infinite polydisk .
Proposition 6.17. If f ∈ H2 and z ∈ D∞ ∩ `2 , then (Qf )(z) is

well-defined.
Proof: Using the Cauchy-Schwarz inequality, we obtain
∞
! ∞ !
X X
|Qf (z)|2 ≤ |an |2 |z|2r(n) .
n=1 n=1
For z ∈ D∞ , observe that the map n 7→ ψz (n) := z [n] is multiplicative

and satisfies |ψz (n)| ≤ 1, for all n ∈ N+ . It follows that
∞ ∞
X r(n) 2
z
Y 1
=
n=1 i=1
1 − |zi |2
Y 1
= 2
.
p∈P
1 − |φ(p)|
Therefore "∞ #1/2
Y 1
|Qf (z)| ≤ kf kH2 .
i=1
1 − |zi |2
This is finite if z ∈ D∞ ∩ `2 .
Remark 6.18. A character on (N+ , ·) is a multiplicative map from

N to T. A quasi-character on (N+ , ·) is a multiplicative map from N+
+
to D. So ψz is a quasi-character.
Hilbert, in 1909, asked:
Question 6.19. Does Qf (z) make sense on a larger set than D∞ ∩
2
`?
This was his answer. Let z = (z1 , z2 , . . . ), and let z(m) denote
(z1 , . . . , zm , 0, 0, . . . ). Consider the sequence Fm (z) := F (z(m) ); this is
called the mte -Abschnitt (or cut-off). If f ∈ H2 and F = Qf , then the
functions Fm are well-defined on D∞ by Proposition 6.17.
Proposition 6.20. (Hilbert) Suppose that there exists C > 0
such that
|Fm (z)| ≤ C ∀ z ∈ D∞ , ∀ m ∈ N+ .
Then, for every z ∈ D∞ ∩ c0 , the limit
lim Fm (z) =: F (z)
m→∞
exists.
6.5. BESICOVITCH’S THEOREM 63
Proof: Fix z ∈ D∞ ∩ c0 and an ε > 0. Then, there exists K ∈ N

ε
such that |zk | < 2C holds for all k > K. Fix n > m > K, and consider
∞
the function f ∈ H (Dn−m ) given by
f (wm+1 , . . . , wn ) := F (z1 , . . . , zm , wm+1 , . . . , wn , 0, 0, . . . ).
Now, we apply the polydisk version of Schwarz’s lemma, Lemma 11.2,
to g(w) := f (w)−f
2C
(0)
. Since g : Dn−m → D, we conclude that
ε
|g(zm+1 , . . . , zn )| ≤ max |zi | < ,
i=m+1,...,n 2C
so that
ε
· 2C = ε.
|f (z) − f (0)| ≤
2C
Thus, the sequence {Fm (z)}m is Cauchy.
Definition 6.21. We define H ∞ (D∞ ) by

X∞
∞ ∞ r(n) ∞
H (D ) := F (z) = an z : |Fm (z)| ≤ C, ∀ m ∈ N, z ∈ D .
n=1
(6.22)
∞ ∞
The norm of F ∈ H (D ) is the smallest C that satisfies the inequality
in (6.22).
6.5. Besicovitch’s Theorem

Definition 6.23. (1) Let f ∈ Hol (Ωρ ), let ε > 0. We say
that τ ∈ R is an ε-translation number of f , if
sup |f (s + iτ ) − f (s)| < ε.
s∈Ωρ
We shall let E(ε, f ) denotes the set of ε-translation numbers

of f .
(2) A set S ⊂ R is called relatively dense, if there exists L <
∞ such that each interval of length L contains at least one
element of S.
(3) A function f ∈ Hol (Ωρ ) is uniformly almost periodic in Ωρ , if
for all ε > 0, the set of ε-translation numbers of f is relatively
dense.
Example 6.24. The function f (s) = 2−s + 3−s is uniformly almost
periodic in the half-plane Ωρ for every ρ ∈ R.
It follows from Kronecker’s theorem that for every ε > 0 there
exists an arbitrarily large ε-translation number. Indeed, let θ1 = log
2π
2
,
θ2 = log
2π
3
and α1 = α2 = 0. Then there exists an arbitrarily large τ ∈ R
so that

τ log 2 τ log 3
dist ,Z < ε and dist , Z < ε.
2π 2π
Thus,
|2−(s+iτ ) − 2−s | = |2−s (e−iτ log 2 − 1)|

−ρ τ log 2
≤ 2 2π(log 2) dist ,Z
2π
< Cε.
Similarly, one obtains
τ log 3
|3−(s+iτ ) − 3−s | ≤ 3−ρ 2π(log 3) dist { , Z} < Cε.
2π
There exists a refined version of Kronecker’s theorem that implies that
the ε-translation numbers of f are relatively dense, so f is uniformly
almost periodic. However, the claim also follows from Corollary 6.28
below.
Theorem 6.25. (Besicovitch) Suppose f (s) = ∞ −s
P
n=1 an n and
the series converges uniformly in Ωρ . Then f is uniformly almost pe-
riodic.
Lemma 6.26. Suppose f is uniformly almost periodic and uniformly
continuous in Ωρ , and let 0 < ε1 < ε2 be arbitrary. Then there exists
a δ > 0 such that for each τ ∈ E(ε1 , f ), the inclusion (τ − δ, τ + δ) ⊂
E(ε2 , f ) holds.
Proof: Let δ > 0 be such that for every 0 < δ 0 < δ and z ∈ Ωρ ,
|f (z + iδ 0 ) − f (z)| < ε2 − ε1 .
For any τ 0 ∈ (τ − δ, τ + δ), write τ 0 = τ + δ 0 with 0 < |δ 0 | < δ. Then
the inequality
|f (z + iτ 0 ) − f (z)| ≤ |f (z + i(τ + δ 0 ) − f (z + iτ )| + |f (z + iτ ) − f (z)|
< (ε2 − ε1 ) + ε1 = ε2
holds.
Lemma 6.27. Let ε, δ > 0 and let f1 , f2 be uniformly almost periodic

and uniformly continuous functions. Then the set
P = {τ ∈ E(ε, f1 ) : dist (τ, E(ε, f2 )) < δ}
is relatively dense.
6.5. BESICOVITCH’S THEOREM 65
Proof: For a uniformly almost periodic function f and ε > 0, let

L(ε, f ) denote the infimum of those L > 0 such that any interval of
length L contains an ε-translation number of f . Choose K ∈ N so that
L = δK is greater than max{L( 2ε , f1 ), L( 2ε , f2 )}. Write
[ [
R = (n − 1)L, nL = In .
n∈Z n∈Z
(n) (n)
In each In there exist τ1 ∈ E( 2ε , f1 ) and ∈ E( 2ε , f2 ) and clearly
τ2
(n) (n)
−L < τ1 − τ2 ≤ L. Decompose [−L, L) into 2K disjoint intervals Jl
of length δ. Since this is a finite number, there exists n0 ∈ N such that
(n) (n)
if any interval Jl contains some point in the set {τ1 − τ2 }n∈Z , then
(n) (n) 0
it contains a point in the set {τ1 − τ2 }nn=−n 0
. Thus, for any n ∈ Z,
0
there exists n ∈ {−n0 , . . . , n0 } such that

(n) (n) (n0 ) (n0 )
(τ1 − τ2 ) − (τ1 − τ2 ) < δ.
Equivalently,

(n) (n0 ) (n) (n0 )
τ := τ1 − τ1 = τ2 − τ2 + θδ,
with |θ| < 1. By the triangle inequality, this implies that τ lies in
(n)
E(ε, f1 ), and is closer than δ to an element of E(ε, f2 ), namely (τ2 −
(n0 )
τ2 ). In other words, τ ∈ P .
We will now show that P is relatively dense. Consider an arbitrary
(n)
interval I of length (2n0 + 3)L and find the integer n for which τ1 is
(n)
closest to the center of I. Then the distance of τ1 from the center of
I is at most L. Find the corresponding n0 and τ , and conclude that
(n) (n0 )
|τ − τ1 | = |τ1 | ≤ n0 L.
This means that τ lies in I, and so the set P intersects every interval
of length (2n0 + 3)L.
Corollary 6.28. Let f1 and f2 be both uniformly almost periodic

and uniformly continuous. Then f1 + f2 is also uniformly almost peri-
odic.
Proof: Fix ε > 0, and apply Lemma 6.26 to f = f2 , ε1 = 3ε and
ε2 = 2ε 3
. We obtain δ > 0 such that {τ : dist (τ, E( 3ε , f2 ) < δ} ⊆
E( 2ε
3
, f2 ). Now apply Lemma 6.27 to conclude that
{τ ∈ E( 3ε , f1 ) : dist (τ, E( 3ε , f2 )) < δ}
is relatively dense. But, by the triangle inequality, any τ in the above
set is an ε-translation number for f1 + f2 .
Proof: (of Theorem 6.25) Since a finite Dirichlet series is uniformly

continuous, it follows inductively from Corollary 6.28 that it is also
uniformly almost periodic. Therefore it is sufficient to prove that the
uniform limit of uniformly almost periodic functions is also uniformly
almost periodic.
Fix ε > 0. Find N so that kfn − f k∞ < ε/3 holds for all n ≥ N .
Then any ε/3-translation number τ of fN is an ε-translation number
of f , since
|f (z + τ ) − f (z)| ≤ |f (z + τ ) − fN (z + τ )| + |fN (z + τ ) − fN (z)| + |fN (z) − f (z)|
< ε ∀ z.
6.6. The spaces Hw2

Definition 6.29. Let w = {wn }∞ n=1 be a sequence of positive real
numbers which are in this context called a weight. Define the Hilbert
space Hw2 of Dirichlet series by
X X
2 −s 2
Hw := an n : |an | wn < ∞ .
n n
Remark 6.30. Note that if f ∈ Hw2 , then f 0 is in the space with

weights wn (log n)2 .
One way to obtain interesting weights is from measures on the pos-
itive real axis. Let µ be a positive Radon measure on [0, ∞) such that
0 ∈ supp µ (6.31)
R∞
0
4−σ dµ(σ) < ∞. (6.32)
We define the weight sequence by
Z ∞
wn := n−2σ dµ(σ). (6.33)
0
One example of course is when µ is the Dirac measure at 0 denoted
by δ0 , and all the weights are 1, giving H2 . Here is another class.
Example 6.33. For each α < 0, define µα on [0, ∞) by
2−α −1−α
dµα (σ) = σ dσ.
Γ(−α)
Then for each n ≥ 2, we have from (6.33)
wn = (log n)α . (6.34)
an n−s
P
Since w1 is infinite, it is convenient to assume that sums n
start at n = 2 when dealing with these spaces.
2
6.6. THE SPACES Hw 67
Remark 6.35. On the unit disk, one can define spaces Hw2 by
X X
2 n 2
Hw := an z : |an | wn < ∞ . (6.36)
n n
A special case is when
wn = (n + 1)α .
Then α = 0 corresponds to the Hardy space, α = −1 to the Bergman
space, and α = 1 to the Dirichlet space, the space of functions whose
derivatives are in the Bergman space. The theory of the Hardy space
on the disk is fairly well-developed – see e.g. [Koo80, Dur70] for a
first course, or [Nik85] for a second. The Bergman space (and the
other spaces with α < 0 in this scale, that all come from L2 -norms
of radial measures) is more complicated — see e.g. [DS04, HKZ00].
The Dirichlet space on the disk is even more complicated analytically,
though it does have the complete Pick property. See e.g. [EFKMR14].
This section should be seen as an attempt to continue the analogy
of Remark 6.35. The case α = 0 in (6.34) we think of as a Hardy-type
space, and the case α = −1 in (6.34) we think of as a Bergman-type
space. When α > 0, we can still define weights by (6.34), though they
do not come from a measure as in (6.33). By Remark 6.30, we can think
of α = 1, for example, as the space of functions whose first derivatives
lie in the space with α = −1. This would render this space a “Dirichlet
space” of Dirichlet series, which is perhaps a surfeit of Dirichlet.
If the weights are defined by (6.33), then, for every ε > 0,
Z ε
wn ≥ n−2σ dµ
0
≥ µ([0, ε])n−2ε , (6.37)
and, consequently, the weight sequence cannot decrease to 0 very fast.
Proposition 6.38. Suppose wn is a weight sequence that is bounded
below by n−2ε for every ε > 0. Then for any f ∈ Hw2 , we have σa (f ) ≤
1
2
.
Proof: Take σ > 12 , and choose ε > 0 such that σ − ε > 12 . Then,
by the Cauchy-Schwarz inequality,
X X
|an |n−σ = |an n−ε |n−(σ−ε)
n n
! 21
X X 21
≤ |an |2 n−2ε n−2(σ−ε) .
n
The first term is finite by (6.37), and the second since 2(σ − ε) > 1.
The following theorem, in the case that µ = δ0 , is due to F. Carlson
[Car22]. If w1 < ∞, we assume that the Dirichlet series for f starts
at n = 1; if w1 is infinite, we start the series at n = 2 (Condition (6.32)
says that w2 < ∞).
Theorem 6.39. Let µPsatisfy (6.31) and (6.32), and define wn by
(6.33). Assume that f = n an n−s has σb (f ) ≤ 0. Then
Z T Z ∞
X
2 1
|an | wn = lim lim |f (s + c)|2 dµ(σ) dt. (6.40)
c→0+ T →∞ 2T −T 0
n
Moreover, if µ({0}) = 0, then the right-hand side becomes

Z T Z ∞
1
lim |f (s)|2 dµ(σ) dt.
T →∞ 2T −T 0
Proof: Fix 0 < c < 1, and let 0 < ε < 1. Define δ by

ε
δ = 1 .
(1 + µ[0, c ])(1 + 2kf kΩc )
Since the Dirichlet series of f converges uniformly in Ωc , there exists
N such that
X
| an n−s − f (s)| < δ, ∀ s ∈ Ωc , ∀ N 0 > N.
n≤N 0
Then
Z T Z 1/c Z T Z 1/c
1 1 X
lim 2
|f (s + c)| dµ(σ) dt = lim | an n−s−c |2 dµ(σ) dt + O(ε)
T →∞ 2T −T 0 T →∞ 2T −T 0 n≤N 0
X Z 1/c
= 2
|an | n−2σ−2c dµ(σ) + O(ε)
n≤N 0 0
Let N 0 tend to infinity, and c tend to 0, to get that the difference

between the left and right sides of (6.40) are at most ε; since this is
arbitrary, the two
R Tsides must 2be equal.
As limT →∞ −−T |f (s + c)| dt is monotonically increasing as c →
0+ , the monotone convergence theorem proves the second part of the
theorem.
In particular, if dµ = dµ−1 = 2dm, we obtain
Z T Z ∞
2 1
X
|an | = 2 lim − |f (s)|2 dm(σ) dt,
n
log n T →∞ −T 0
6.7. MULTIPLIER ALGEBRAS OF H2 AND Hw
2
69
and for µ = δ0 , we get

X Z T
|an | = lim lim − |f (c + it)|2 dt.
2
c→0+ T →∞ −T
n
6.7. Multiplier algebras of H2 and Hw2

Notation 6.41. Let us denote by D the set of functions expressible
as Dirichlet series which converge somewhere, that is,
∞
X
an n−s in Ωρ .

D := f : ∃ρ such that f (s) =
n=1
Since σa ≤ σc + 1, D is also the set of Dirichlet series that converge

absolutely in some half-plane.
The following theorem is due to H. Hedenmalm, P. Lindqvist and
K. Seip, in their ground-breaking paper [HLS97].
Theorem 6.42. Let µ and {wn } satisfy (6.31) – (6.33). Then
Mult (Hw2 ) is isometrically isomorphic to H ∞ (Ω0 ) ∩ D.
Remark 6.43. Before we prove the theorem, note that it implies

that the multiplier algebra is independent of the weight w. The sit-
uation is analogous to a similar phenomenon on the disk. For any
sequence w = {wn }∞ n=0 , one can define a Hilbert space of holomorphic
functions Hw2 by (6.36). If the sequence w comes from a radial positive
Radon measure µ on D such that T ⊂ supp µ as
Z
wn = |z|2n dµ(z),
D
then {wn }n is non-increasing and, since the measure is radial, the se-
quence {z n }n∈N is an orthogonal basis of Hw2 . (Saying the measure is
radial means dµ = dθdν(r) for some measure ν on [0, 1]). Thus, the
norm on Hw2 is given by integration:
Z
2
kf k = |f (z)|2 dµ(z).
D
For all these spaces,

Mult (Hw2 ) = H ∞ (D), (6.44)
the bounded analytic functions on the disk. Indeed, if µ is carried by
the open disk, this follows from Proposition 11.9. If µ puts weight on
the circle, the theorem is still true, and can most easily be seen by
writing Z Z
|f (z)| dµ(z) = lim |f (rz)|2 dµ(z).
2
D r%1 D
In particular, (6.44) holds for all the spaces with wn = (n + 1)α for
α ≤ 0.
Remark 6.45. There exist many functions in H ∞ (Ω0 ) \ D, for
−s
example f (s) = 23 s
and g(s) = (s+1)2.
Before embarking on the proof of the theorem, recall the following

fact. It is a version of the Phragmén-Lindelöf principle — a maximum
modulus principle for unbounded domains. This particular version is
known as the three line lemma.
Lemma 6.46. Let f be a bounded holomorphic function in {z ∈
C; a < Re z < b}, let N (σ) := supt∈R |f (σ + it)|. Then the function
N is logarithmically convex, that is,
b−σ σ−a
N (σ) ≤ N (a) b−a N (b) b−a .
Proof: See Theorem 12.8, p. 274 in [Rud86].
Remark 6.47. The lemma does not hold without the assumption
that f is bounded in the strip. Indeed, consider the function f (z) =
iz
ee . It is holomorphic in the strip {− π2 < Re z < π2 }, bounded on
its boundary {|Re z| = π2 }, but limt→−∞ f (it) = ∞. However, one
can weaken the assumption of boundedness of f to an appropriate
restriction on the growth of f .
The following lemma is trivial if 1 ∈ Hw2 .
Lemma 6.48. Any multiplier of Hw2 lies in D.
Proof: If ϕ belongs to Mult (Hw2 ), then both ϕ(s)2−s and ϕ(s)3−s
are in D. So
X
ϕ(s)2−s = an n−s
X
ϕ(s)3−s = bn n−s .
Multiplying the first equation by 3−s and the second by 2−s , we con-
clude that an is zero when n is odd (and bn is zero when n is not
divisible by 3), so ϕ itself can be represented by an ordinary Dirichlet
series.
P∞ −s
Proposition 6.49. Let ϕ(s) = n=1 bn n with σb ≤ 0. Then
kMϕ k = kϕkΩ0 .
2
71
an n−s , then σb (ϕf ) ≤ 0. By Theorem

P
Proof: Let f (s) = n≤N
6.39,
Z T Z ∞
kϕf k2Hw2 = lim lim − |ϕ(s + c)|2 |f (s + c)|2 dµ(σ) dt
c→0+ T →∞ −T 0
≤ kϕk2Ω0 · kf k2Hw2 .
Hence Mϕ is bounded on a dense subset of Hw2 , and therefore extends
to a bounded operator on all of Hw2 , which must be multiplication by
φ. (Why?) Also, the estimate above shows that kMϕ k ≤ kϕkΩ0 .
Conversely, assume that kMϕ k = 1 and 1 < kϕkΩ0 (possibly infi-
nite). Let
N (σ) := sup |ϕ(σ + it)|.
t∈R
Clearly, N (σ) → |b1 | as σ → ∞, and for any σ > 0, we have

Z T X∞
2
N (σ) ≥ lim − |ϕ(σ + it)| dt = 2
|bn |2 n−2σ > |b1 |2 ,
T →∞ −T n=1
unless ϕ is a constant (in which case the Proposition is obvious). For

any 0 < a < b one can apply the three line lemma, 6.46, to conclude
that log N is convex, so it must be convex on the half-line (0, ∞). Since
lim log N (σ) = log |b1 | < ∞,
σ→∞
we must have that log N , and hence N , is a decreasing function on

(0, ∞).
For each c > 0, n bn n−s converges uniformly in Ωc , and hence by
P
Theorem 6.25, ϕ is uniformly continuous and uniformly almost periodic
in this half-plane. Thus, there exist ε1 , ε2 , ε3 and ε4 positive such that

{t : |ϕ(σ + it)| ≥ 1 + ε1 , −T < t < T } ≥ ε2 (2T ) (6.50)
holds for every sufficiently large T > 0, and σ ∈ (ε3 , ε3 + ε4 ). Indeed,
choose ε3 so that N (ε3 ) > 1. Then there is some ε1 > 0 and some
rectangle R with non-empty interior,
R = {σ + it : ε3 ≤ σ ≤ ε3 + ε4 , t1 ≤ t ≤ t1 + h},
such that |ϕ| > 1 + 2ε1 on R. By the definition of uniform almost
periodicity, there exists some L such that every interval of length L
contains an ε1 translation number of ϕ. For T > L, every interval of
length 2T contains at least TL disjoint sub-intervals of length L, so for
any σ ∈ [ε3 , ε3 + ε4 ] the left-hand side of (6.50) is at least TL h. Setting
h
ε2 = 2L yields the inequality (6.50).
Now, on one hand, we have

kMϕj 2−s kHw2 ≤ kMϕ kj · k2−s kHw2 = k2−s kHw2 ,
√
so that this sequence of norms is bounded by w2 . On the other hand,
Z ε4 Z T
j −s 2
kMϕ 2 kHw2 ≥ lim − |2−(s+ε3 ) ϕj (s + ε3 )|2 dt dµ(σ)
T →∞ 0 −T
≥ µ([0, ε4 ])2−2(ε3 +ε4 ) ε2 (1 + ε1 )2j ,
and this tends to infinity as j tends to ∞, a contradiction.
For later use, note that the proof of Proposition 6.49 shows:
Lemma 6.51. If ϕ = ∞ −s
P
n=1 bn n satisfies σb (ϕ) ≤ 0, and kϕkΩ0 >
1, then
sup kMϕj 2−s k = ∞.
j∈N+
For K ∈ N+ , define
NK = {n = pr11 · . . . · prKK ; rj ∈ N},
where,
as usual, pl is the l-th prime. Clearly, n ∈ NK , if and only if
pl 6 n for all l > K. Let QK : D → D be the map defined by

X ∞ X
−s
QK an n = an n−s .
n=1 n∈NK
P∞
The map QK is well-defined, since if n=1 an n−s converges absolutely
in Ωρ , then so does n∈NK an n−s .
P
We need the following observations.
Lemma 6.52. For any K ∈ N+ , the map QK has the following
properties:
(1) The restriction of QK to Hw2 is the orthogonal projection onto
span {n−s : n ∈ NK }.
(2) For any ϕ, f ∈ D, QK (ϕf ) = (QK ϕ)(QK f ).
(3) If ϕ ∈ Mult (Hw2 ), then QK Mϕ QK = MQK ϕ QK = QK Mϕ .
Proof: (1) follows immediately from the orthogonality of the func-
tions {n−s }n∈N+ .
(2) By linearity, we only need to check that QK (n−s m−s ) =
QK (n−s )QK (m−s ), for +
all m, n ∈ N . This follows from the facts that
if p is prime, then p6 nm if and only if p6 n and p6 m, and
(
−s n−s , pl 6 n, for all l > K,
QK n =
0, otherwise.
2
73
(3) Let f ∈ Hw2 , then, using (2), we get

QK Mϕ QK f = QK (ϕQK f )
= (QK ϕ)(Q2K f )
= QK (ϕf ).
Also,
MQK ϕ QK f = (QK ϕ)(QK f )
= QK (ϕf ).
Proposition 6.53. Mult (Hw2 ) ⊂ H ∞ (Ω0 ) ∩ D.
Proof: Let f = an n−s ∈ Hw2 , and fix K ∈ N+ , s ∈ Ω0 . Then
P

X
|QK f (s)| = an n−s

n∈NK
" #
X
≤ n−σ sup |an |
n∈NK
n∈NK
" K
#
Y 1
= sup |an |.
j=1
1 − p−σ
j n∈NK
supn∈NK |an | is finite, then QK (f ) is bounded in Ωρ for all ρ > 0.

So, if P
Since n |an |2 ωn converges, {|an |2 ωn } is bounded, and hence by (6.37),
|an | = O(nε ) for all ε > 0. Thus, for any ε > 0, the Dirichlet series
of fε (s) := f (s + ε) has bounded coefficients. Consequently, QK fε ∈
H ∞ (Ωρ ), which is the same as saying QK fε+ρ ∈ H ∞ (Ω0 ). Since ε > 0
and ρ > 0 were arbitrary, we conclude that
QK fε ∈ H ∞ (Ω0 ), ∀ K ∈ N+ , ε > 0, f ∈ Hw2 .
Let ϕ be in Mult (Hw2 ). Then ϕ2−s ∈ Hw2 , and so 2−s QK (ϕ) =
QK (ϕ2−s ) ∈ H ∞ (Ωε ), for all ε > 0. Since we know ϕ ∈ D by
Lemma 6.48 it follows that σb (QK ϕ) ≤ 0.
By Lemma 6.51, applied to QK ϕ, we get
kQK ϕkΩ0 ≤ kMQK ϕ |QK Hw2 k. (6.54)
By Lemma 6.52,
kMQK ϕ |QK Hw2 k = kQK Mϕ QK k
≤ kMϕ k.
So by (6.54),
kQK ϕkΩ0 ≤ kMϕ k ∀ K ∈ N+ .
Using normal families, we conclude that some subsequence QKl ϕ con-

verges to some function ψ ∈ H ∞ (Ω0 ) uniformly on compact subsets of
Ω0 . But QK ϕ → ϕ uniformly on compact subsets of Ωσu (ϕ) and hence,
ϕ = ψ in Ωσu (ϕ) . By uniqueness of analytic functions, we conclude that
ϕ = ψ in Ω0 .
Combining Propositions 6.49 and 6.53, we complete the proof of
Theorem 6.42. This also concludes the solution to Beurling’s problem
[HLS97].
√ P Corollary 6.55. (Hedenmalm, Lindqvist, Seip) Let ψ(x) =

2 ∞ n=1 cn sin(nπx) be an odd, 2-periodic function on R. Then
{ψ(nx)}n∈N+Pforms a Riesz basis for L2 ([0, 1]) if and only if the func-
tion ϕ(s) = ∞ n=1 cn n
−s
is bounded and bounded from below in Ω0 .
6.8. Cyclic Vectors

Consider the following variant of Beurling’s question. Let ψ :
[0; 1] → C be in L2 . When is the set {ψ(nx) : n ∈ N+ } complete,
i.e., when do we have
span {ψ(nx) : n ∈ N+ } = L2 ([0; 1])?
ψ(x) = ∞
P
As before, we can writeP n=1 cn β(x), and translate this prob-
2 ∞ −s
lem to H . Let f (s) = n=1 cn n . When is
span {f (ns) : n ∈ N+ } = H2 ?
Since f (ns) = (Mn−s f )(s), it is equivalent to requiring that
span {f · D} = H2 ,
i.e. that f is a cyclic vector for the collection of multipliers {Mp−s :
p ∈ P}. An obvious necessary condition is that f does not vanish in
Ω1/2 . We record this open question.
Question 6.56. Which Dirichlet series f satisfy span {f ·D} = H2 ?
6.9. Exercises
Exercise 6.57. Show that H2 contains a function f with σa (f ) =
1
2
.
Exercise 6.58. Prove that the reproducing kernel for Hw2 is given
by
X 1
k(s, u) = n−s−ū . (6.59)
n
w n
Exercise 6.60. Prove (6.34).

6.10. NOTES 75
Exercise 6.61. Show that if α ∈ Z, and wn = (log n)α , the re-

producing kernel for Hw2 can be written in terms of the ζ function (if
α = 0), its derivatives (if α < 0) or integrals (if α > 0), after adjusting
if necessary for the constant term.
an n−s is in D if and only if an is
P
Exercise 6.62. Prove that
bounded by a polynomial in n.
6.10. Notes
The proof we give of Besicovitch’s theorem 6.25 is from his book
[Bes32, p. 144]. In the book he also develops the theory of functions
that are almost periodic in the Lp -sense (where the Lp -norm of the
difference between f and a vertical translate of it is less than ).
The solution to Beurling’s problem, and the proof of Theorem 6.42
(in the most important case, Hw2 = H2 ) is due to Hedenmalm, Lindqvist
and Seip [HLS97]. The spaces Hw2 were first studied in [McCa04].
In Carlson’s theorem 6.39, if µ has a point mass at 0, then one
cannot take the limit with respect to c inside the integral in (6.40).
Indeed, E. Saksman and K. Seip prove the following theorem in [SS09]:
Theorem 6.63. (1) There exists a function f in H ∞ (Ω0 ) ∩ D such
1
RT
that limT →∞ 2T −T
|f (it)|2 dt does not exist.
(2) For all ε > 0, there exists g = ∞ −s ∞
P
Pan n 2 ∈ H (Ω0 ) ∩ D that
n=1
is a singular inner function and such that |an | < ε.
For a more refined version of Carlson’s theorem, see [QQ13, Section
7.4].
CHAPTER 7
Characters
7.1. Vertical Limits

Let us return to the map Q : D → Hol (D∞ ). Consider the group
+
(Q , ·) equipped with discrete topology. Its dual group K — the group
of all characters,
K = {χ : Q+ → T; χ(mn) = χ(m)χ(n), for all m, n ∈ Q+ }
is isomorphic (as a topological group) to T∞ via the map χ 7→
{χ(pk )}k∈N+ = (χ(2), χ(3), χ(5), . . . ). The topology on K is the topol-
ogy of pointwise convergence. It corresponds to the product topology
on T∞ . The group T∞ is also equipped with a Haar measure, which
is the infinite product of the Haar measures on T. We shall use ρ to
denote Haar measure on T∞ .
Given any set X, a flow on X is family of maps Tt : X → X, where
t is a real parameter, that satisfy T0 is the identity, and Ts ◦Tt = Ts+t . If
X is equipped with some structure (measure space, topological space,
smooth manifold, ...), we usually assume that Tt is compatible with
this structure (i.e. each Tt is measurable, continuous, smooth, ...).
Given a sequence of real number {αn }n∈N , we define a flow on T∞
by
Tt (z1 , z2 , . . . ) := (e−itα1 z1 , e−itα2 z2 , . . . ),
the so-called Kronecker flow . Note that the Kronecker flow is contin-
uous and measurable.
Definition 7.1. A measurable flow on a probability space is er-
godic, if all invariant sets have measure 0 or 1.
Theorem 7.2. The Kronecker flow is ergodic if and only if {αn }

are linearly independent over Q.
Proof: See [CFS82].
In particular, if αn = log pn , the Kronecker flow is ergodic. (See
Theorem 6.14.) The ergodic theorem (of which there are many vari-
ants) says that for an ergodic flow, the time average (the left-hand side
of (7.4)) equals the space average (the right-hand side).
77
78 7. CHARACTERS
Theorem 7.3. (Birkhoff-Khinchin) Let Tt be an ergodic flow on

K. Then
Z T Z
1
lim g(Tt χ0 ) dt = g(χ)dρ(χ), (7.4)
T →∞ 2T −T K
for all χ0 , if g ∈ C(K), and for a.e. χ0 , if g ∈ L1 .

Proof: See [CFS82].
P∞
Lemma 7.5. Let f ∼ n=1 an n−s safisfies σu (f ) < 0. Then Qf ∈
C(T∞ ).
Proof: It suffices to show that the series for Qf is uniformly Cauchy,
since the partial sums are clearly continuous.
Let L = supn |an | < ∞. Fix 0 < ε < 1, and find N ∈ N such that
for all M2 > M1 > N
M2
X
it

an n < ε.
n=M1
Thus, for all t ∈ R,

M2
X it log p r1 (n) it log p rk (n)

an e 1
... e k < ε.

n=M1
Note that, if w1 , . . . , wk , ζ1 , . . . , ζk ∈ T, then

|w1 . . . wk − ζ1 . . . ζk | ≤ |w1 − ζ1 | + · · · + |wk − ζk |. (7.5)
This can be proven by induction on k using the inequality |w1 w2 −
ζ1 ζ2 | ≤ |w1 − ζ1 | + |w2 − ζ2 |, which follows easily from the triangle
inequality.
Fix z ∈ T∞ and M2 > M1 > N as above. By Kronecker’s theorem
6.14, we can find t ∈ R such that |eit log pj − zj | < Mε2 L holds for all j’s
such that pj ≤ M2 . Thus we have,
M2 M2 h
X
r(n)
X r(n)
it log p1 r1 (n) it log p rk (n) i

an z ≤ an z − e ... e k

n=M1 n=M1
M2
X it log p r1 (n) it log p rk (n)
+
an e 1
... e k

n=M1
< ε + ε = 2ε,
where we used the inequality (7.5) to estimate the first term.
This gives another proof of Carlson’s theorem, 6.39.
7.1. VERTICAL LIMITS 79
Theorem 7.6. Let f ∼ ∞ −s

∈ H2 , and let x > 21 . Then
P
n=1 an n
Z T ∞
1 X
lim 2
|f (x + it)| ds = |an |2 n−2x . (7.7)
T →∞ 2T −T
n=1
Proof: Since σu (f ) ≤ σa (f ) ≤ 21 , we obtain σu (fx ) < 0, for x > 12 .

Since Qfx is continuous on T∞ by Lemma 7.5, we can apply the
Birkhoff-Khinchin ergodic theorem 7.3 for any character χ0 ∈ K to get
Z T Z
1 2
lim |Qfx (Tt χ0 )| dt = |Qfx (χ)|2 dρ(χ)
T →∞ 2T −T K
X
= dx (q)|2
|Qf (7.8)
q∈Q+
∞
X
= |an |2 n−2x . (7.9)
n=1
We used Plancherel’s theorem to obtain (7.8), and the fact that Qf is

a sum only over positive powers of z means the only non-zero terms
in (7.8) are when q ∈ N+ , giving (7.8). Choosing the trivial character
χ0 (n) ≡ 1 yields
X
(Qfx )(Tt χ0 ) = an n−x n−it χ0 (n)
n
= f (x + it),
giving (7.7).
For every τ ∈ R, the map f 7→ fiτ is unitary on H2 . Thus, by
Corollary 11.8, for every sequence {τk }k∈N ⊂ R, there is a subsequence
τkl such that {fiτkl } converges uniformly on compact subsets of Ω1/2 .
Definition 7.10. Let f ∈ H2 , and let {τk }k∈N be a sequence of real
numbers. If the sequence fiτk converges uniformly on compact subsets
of Ω1/2 to a function g, then g is called a vertical limit function of f .
Proposition
P∞ 7.11. Let f ∈ H2 , and let χ be a character. Then
fχ (s) := n=1 an χ(n)n−s is a vertical limit function of f . Conversely,
all vertical limit functions have this form for some character χ.
Proof: Fix a character χ and let k ∈ N+ . By Kronecker’s theorem,
we can find τk ∈ R such that |eiτk log pj − χ(pj )| ≤ 1/k holds for j =
1, . . . , k. Define fk := fiτk . Then using inequality (7.5), we conclude
r (n) r (n)
that for any n ∈ N+ , n = p11 . . . pl l ,
fbk (n) − fbχ (n) = fˆ(n)niτk − fˆ(n)χ(n)

80 7. CHARACTERS
l l
Y i log pj τ rj Y
fˆ(n)·

rj
= e k
− χ(p j )
j=1 j=1
l
X
≤ kf kH2 rj ei log pj τk − χ(pj )
j=1
l
1 X
≤ kf kH2 rj ,
k j=1
and this last expression tends to 0 as k → ∞. Proposition 11.7 now

implies that fχ is a vertical limit function of f .
Conversely, let g be a vertical limit function of f . Using Proposition
11.7 again, we conclude that there exists a sequence {τk }k∈N ⊂ R such
that fbiτk (n) → ĝ(n) for all n ∈ N. Equivalently,
ĝ(n)
niτk → , as k → ∞.
fˆ(n)
Since n 7→ niτk is a character for all k ∈ N, so is the limit: n 7→

ĝ(n)/fˆ(n).
Let us now turn to the Lindelöf hypothesis, a conjecture weaker

than the Riemann hypothesis, but one that could be possibly ap-
proached by the tools of functional analysis.
Recall that the alternating zeta function is given by ζ̃(s) =
P∞ n −s 1−s
n=1 (−1) n . We have seen that ζ̃(s) = (2 − 1)ζ(s). This im-
plies that ζ̃(s) and ζ(s) are of comparable size in {s ∈ C : Re s >
0, |1 − Re s| > ε}, for any ε > 0.
1
Conjecture 7.12. (Lindelöf hypothesis) For every σ > 2
and
k ∈ N+
Z T
1
lim |ζ̃ k (σ + it)|2 dt < ∞
T →∞ 2T −T
holds.
Recall that dk (n), defined in Corollary 1.17, is the number of ways

n can be factored into exactly k factors, allowing 1 and where the order
matters.
Lemma 7.13. Let k be a natural number and let ε > 0. Then
dk (n) = O(nε ) as n → ∞.
7.1. VERTICAL LIMITS 81
Proof: Note that d2 (n) is the number of divisors of n. Also, d3 (n) ≤

d2 (n)2 , since
X n X
d3 (n) = d2 ≤ d2 (n) = d2 (n)2 .
l
l|n l|n
Applying this argument inductively, we obtain dk (n) ≤ d2 (n)k−1 and

thus it is enough to show that d2 (n) = O(nε ) for all ε > 0.
Fix ε > 0. We need to show that there exist C = C(ε) such that
d2 (n) ≤ Cnε holds for all n ∈ N+ , or equivalently, that
log d2 (n) ≤ ε log n + log C.
t
Write n = lj=1 pjj with tj ≥ 0 and tl > 0, then d2 (n) = lj=1 (1 + tj ).
Q Q
We want to show that
l
X
[log(1 + tj ) − εtj log pj ] ≤ log C
j=1
for all n ∈ N. Clearly, if log pj ≥ 1/ε, then the j th summand is non-

positive, because log(1 + tj ) < tj . As tj → ∞, the j th summand tends
to −∞. Hence each of the finitely many summands with log pj < 1/ε
is bounded.
Suppose that the Carlson theorem applied to ζ̃ k (s). Then
Z T ∞
1 X
lim k 2
|ζ̃ (σ + it)| dt = n−2σ |ζb̃k (n)|2
T →∞ 2T −T n=1
X∞
≤ n−2σ |ζbk (n)|2
n=1
< ∞,
since ζbk (n) = dk (n) = O(nε ) for all ε > 0 by Lemma 7.13. Thus we
would have proved the Lindelöf hypothesis. Conversely, the following
is known.
Theorem 7.14. (Titchmarsh) If
Z T
1
lim |ζ̃ k (σ + it)|2 dt < ∞,
T →∞ 2T −T
P∞
then it equals to n=1 n−2σ |ζb̃k (n)|2 .
82 7. CHARACTERS
7.2. Helson’s Theorem

We will need some properties of Hardy spaces of the right half-plane
Ω0 . There is more than one natural definition. We will consider two
of them. Let ψ : Ω0 → D be the standard conformal mapping of the
1−z
right half-plane onto the disk, that is, ψ(z) = 1+z . For 1 ≤ p ≤ ∞, we
define the conformally invariant Hardy space as
Hip (Ω0 ) = {g ◦ ψ; g ∈ H p (D)}.
1+it
For 1 ≤ p < ∞, writing eiθ = ψ(−it) = 1−it
, and changing variables
yields
Z
dθ
kgkpH p (D) = |g(eiθ )|p
2π
ZT
p dθ
dt
= |(g ◦ ψ)(−it)|
dt 2π
ZR
dt
= |(g ◦ ψ)(−it)|p .
R π(1 + t2 )
Any function g ∈ H p (D) extends to an Lp function on T satisfying
Z
dθ
g(eiθ )einθ = 0, for all n ∈ N+ . (7.10)
T 2π
Conversely, any Lp function on T satisfying (7.10) is the boundary
value of function in H p (D).
dt
Let µ be the measure on the real axis give by dµ(t) = π(1+t 2) .
We deduce that a Lebesgue measurable function f : iR → C belongs

to Hip (Ω0 ), if and only if
Z
p
kf kH p (Ω0 ) := |f (it)|p dµ(t) < ∞,
i
R
and n
1 − it
Z
f (it) dµ(t) = 0, for all n ∈ N+ . (7.11)
R 1 + it
Here is the second definition for the Hardy spaces of the half-plane.
For 1 ≤ p < ∞, set
Z ∞
p
p

|f (σ + it)|p dt < ∞ .

H (Ω0 ) := f ∈ Hol (Ω0 ) kf kH p (Ω0 ) := sup

σ>0 −∞
For any function f ∈ H p (Ω0 ) and almost every t ∈ R, the limit f˜(it) :=
limσ→0+ f (σ + it) exists and satisfies f˜ ∈ Lp (iR). One can recover f
from f˜ by convolution with the Poisson kernel. For both Hip (Ω0 ) and
H p (Ω0 ) we identify the functions with their boundary values.
7.2. HELSON’S THEOREM 83
By the Paley-Wiener theorem,
H 2 (Ω0 ) = {f ∈ L2 (iR); (Ff )(iξ) = 0, ∀ξ < 0}.
The different integrability conditions in Hip (Ω0 ) and H p (Ω0 ) yield
f (z)
f ∈ Hi2 (Ω0 ) ⇐⇒ ∈ H 2 (Ω0 ).
1+z
When p = ∞, we will define H ∞ (Ω0 ) = Hi∞ (Ω0 ) to be the bounded

analytic functions in Ω0 . For more information on H p spaces of the
half-plane see Chapters 10 and 11 of [Dur70].
Theorem 7.15. (Helson) Let f (s) ∼P ∞ −s

∈ H2 . For a.e.
P
n=1 an n
∞ −s
character χ ∈ K, the function fχ (s) = n=1 an χ(n)n defined on
2
Ω1/2 extends to an element of Hi (Ω0 ), and satisfies
Z T ∞
1 2
X
lim |fχ (it)| dt = |an |2 < ∞. (7.16)
T →∞ 2T −T n=1
Before we prove the theorem, let us start with a preliminary obser-

vation. The set {eq }q∈Q+ forms an orthonormal basis of L2 (K), where
eq (χ) = χ(q), for all χ ∈ K.
Include reference here.

Proof: By Tonelli’s theorem, we have,
Z Z ∞
|fχ (it)|2 dµ(t)dρ(χ)
K −∞
Z ∞Z
= |fχ (it)|2 dρ(χ)dµ(t)
Z−∞∞ Z X
K
n −it
= an am χ(n)χ(m) dρ(χ)dµ(t)
−∞ K m,n m
Z ∞X
= |an |2 dµ(t)
−∞
X n
= |an |2
n
= kf k2H2 .
We conclude that for a.e. χ ∈ K, the function fχ belongs to L2 (iR, dµ).

84 7. CHARACTERS
Fix k ∈ N+ . By the Cauchy-Schwarz inequality and Tonelli’s theo-

rem, we obtain
Z Z k 2
1 − it
fχ (it) dµ(t) dρ(χ)

K R 1 + it
1 − it 2n
Z Z Z
2
≤ |fχ (it)| dµ(t) 1 + it dµ(t) dρ(χ)

K
Z Z R R
= |fχ (it)|2 dρ(χ)dµ(t)

ZR ZK X n −it
= an am χ(n)χ(m) dρ(χ)dµ(t)
R K m,n m
Z X
= |an |2 dµ(t)
R
Xn
= |an |2 < ∞.
n
1−it k
R
Thus, the function G(χ) := R fχ (it)( 1+it ) dµ(t) belongs to L2 (K) ⊂
L1 (K). To show G(χ) = 0 for a.e. χ, we only need to show that all its
Fourier coefficients vanish, i.e,
Z
G(χ)χ(q) dρ(χ) = 0, for all q ∈ Q+ .
K
Let us set aq = 0 for all q ∈ Q+ \ N+ . Since G ∈ L1 (K), we can apply

Fubini’s theorem:
Z k
1 − it
Z Z
G(χ)χ(q) dρ(χ) = χ(q) fχ (it)dµ(t)dρ(χ)
K K R 1 + it
Z k Z
1 − it X
= χ(q) an n−it χ(n)dρ(χ)dµ(t)
R 1 + it K n
Z k
1 − it
= aq q −it dµ(t).
R 1 + it
If q ∈ Q+ \ N+ , then the last term vanishes, since aq does. If q ∈ N+ ,

then q −it ∈ H ∞ (Ω0 ) = Hi∞ (Ω0 ) and thus has the form g ◦ ψ for some
g ∈ H ∞ (D) ⊂ H 2 (D). Hence, by (7.11) the last term above also
vanishes. Consequently, G(χ) = 0 a.e., and so for a.e. χ ∈ K, fχ
belongs to Hi2 (Ω0 ).
To prove (7.16), note that, by Plancherel’s
P∞ theorem,Pthe function
∞
Qf : K → C defined by (Qf )(χ) = n=1 an χ(n) = n=1 an en (χ)
7.2. HELSON’S THEOREM 85
belongs to L2 (K). Also note that

X
fχ (it) = an χ(n)n−it
n
X
= an (Tt χ)(n)
n
= (Qf )(Tt χ),
where Tt is the Kronecker flow on K. We apply the Birkhoff-Khinchin
erdodic theorem 7.3 to the ergodic flow {Tt } and the function |Qf |2 ∈
L1 (K) to conclude that
Z T Z T
1 2 1
lim |fχ0 (it)| dt = lim |(Qf )(Tt χ0 )|2 dt
T →∞ 2T −T T →∞ 2T −T
Z
= |(Qf )(χ)|2 dρ(χ)
K
∞
X
= |an |2 ,
n=1
holds for a.e. χ0 ∈ K.
k
Remark 7.17. Recall that by Lemma 7.13, ζ1/2+ε and, conse-
k
quently, ζ̃1/2+ε belong to H2 for every k ∈ N+ and ε > 0. Thus,
by Theorem 7.15, for a.e. χ ∈ K
Z T 2k
1
lim − ζχ
+ ε + it dt < ∞,
T →∞ −T 2
and
Z T 2k
ζ̃χ 1 + ε + it dt < ∞.

lim −
T →∞ −T
2
A sequence {an }∞ n=1 is called totally multiplicative, if an am = anm

holds for all n, m ∈ N+ .
Lemma 7.18. P Let {an } be a non-trivial totally
P∞ multiplicative se-
quence. If f (s) = ∞ a
n=1 n n −s
, then 1/f (s) ∼ a
n=1 n µ(n)n −s
, where
µ denotes the Möbius function.
Proof: Using Corollary 1.18 we obtain
 
∞
! ∞ ! ∞
X X X X
an n−s am µ(m)m−s = k −s  an ak/n µ(k/n)
n=1 m=1 k=1 n|k
86 7. CHARACTERS
 
∞
X X
= ak k −s  µ(k/n)
k=1 n|k
= a1
= 1.
2
Theorem 7.19. If a sequence P∞ {an } ∈ ` −sis totally multiplicative,
then for a.e. character χ ∈ K, n=1 an χ(n)n extends analytically to
a zero-free function in Ω0 .
P∞ −s
P∞Proof: Write −s
f (s) = n=1 an n and, note that fχ =
n=1 an χ(n)n also has totally multiplicative coefficients. Thus
∞
1 X
= an χ(n)µ(n)n−s = gχ (s),
fχ (s) n=1
P∞ −s
where g(s) = n=1 an µ(n)n ∈ H2 , since µ(n) ∈ {0, ±1} for all
n ∈ N+ . By Theorem 7.15, the function gχ belongs to Hi2 (Ω0 ) for a.e.
χ. Consequently, fχ must be zero-free in the right half-plane for the
same χ’s.
We obtain the following “probabilistic version” of the Riemann hy-
pothesis.
Corollary 7.20. (Helson) For almost every character χ ∈ K,
ζχ (s) = ∞ −s
P
n=1 χ(n)n is zero-free in Ω1/2 .
P Proof: Since {n−(1/2+ε) } ∈ `2 , we conclude that

−(1/2+ε)
nn χ(n)n−s is zero-free in Ω0 for a.e. χ. In other words, ζχ is
zero-free in Ω1/2+ε for a.e. χ. Taking ε = m1 and intersecting the sets
of corresponding χ’s we conclude that ζχ is zero-free in Ω1/2 for a.e. χ.
7.3. Dirichlet’s theorem on primes in arithmetic progressions

We can write the set of all primes P as the disjoint union P =
P0 ∪ P1 ∪ P2 , where
Pj = {p ∈ P; p ≡ j mod 3}
for j = 0, 1, 2.
Clearly, P0 = {3} and the following easy argument shows that P2
is infinite.
Proof: Suppose not and write P2 \ {2} = {q1 , . . . , qN }. Let
M = 3q1 . . . qN + 2.
7.3. DIRICHLET’S THEOREM ON PRIMES IN ARITHMETIC PROGRESSIONS87
Then M ≡ 2 mod 3 and M is not divisible 2 nor by any qj . Thus M

factors as M = q̃1 . . . q̃k with q̃j ∈ P1 for all j. This implies M ≡ 1
mod 3, a contradiction.
It seems that no similar simple argument exists for P1 . Nevertheless,
even more is true: every arithmetic progression without a common
factor contains a set of primes whose reciprocals are not summable.
Theorem 7.21. (Dirichlet, 1837) Let l, q ∈ N+ and assume that
gcd(l, q) = 1. Then
X 1
= ∞.
p∈P; p≡l mod q
p
Before we prove this theorem, we need some preparation. Let q

be a natural number, and let us denote by Z∗q the group of units of
the ring Zq , that is, the group of invertible elements of Zq . It can be
checked that 0 ≤ k ≤ q − 1 is a unit in Zq if and only if gcd(k, q) = 1
(see Exercises ??), and so |Z∗q | = φ(q).
Let G be a finite abelian group, and let `2 (G) be the Hilbert space
of functions f : G → C normed by
1 X
kf k2 := |f (g)|2 .
|G| g∈G
The dual group of G, denoted by Ĝ, is the set of characters, i.e. the
multiplicative functions from G to T.
Proposition 7.22. Let G be a finite abelian group. Then Ĝ forms
an orthonormal basis of `2 (G).
Fix q, and let G := Z∗q . Any character e ∈ Ĝ = Z
c∗ extends to Z by
q
(
e(n mod q), if gcd(n, q) = 1,
e(n) =
0, otherwise.
Then e : Z → T ∪ {0} is totally multiplicative. Any such function is
called a Dirichlet character modulo q. We denote the set of all Dirichlet
characters modulo q by Xq . The trivial Dirichlet character modulo q
is the periodic extension of the trivial character on Z∗q , that is,
(
1, if gcd(n, q) = 1,
χ0 (n) =
0, otherwise.
We will identify Dirichlet characters modulo q with their restriction to
Z∗q .
88 7. CHARACTERS
Suppose that gcd(l, q) = 1 and define δl : Z → T ∪ {0} by

(
1, n ≡ l mod q,
δl (n) =
0, otherwise.
Then δl is q-periodic (but not multiplicative). We can regard it also as
an element of `2 (Z∗q ), and by expansion with respect to the orthonormal
basis obtain consisting of characters
X
δl (n) = hδl , χiχ(n),
χ
if gcd(n, q) = 1. If gcd(n, q) 6= 1, the equality also holds, since both

sides vanish.
P Let Re −s s > 1, then for any Dirichlet character, the series
p∈P χ(p)p converges absolutely. This justifies exchanging the or-
der of the sums in the following
X 1 X δl (p)
=
p≡l mod q; p∈P
ps p∈P
ps
1 XX
= χ(l)χ(p)p−s
φ(q) p∈P χ∈X
q
1 X X χ(p)
= χ(l)
φ(q) χ∈X p∈P
ps
q
" #
1 X χ0 (p) X X χ(p)
= + χ(l) (7.17)
φ(q) p∈P ps χ6=χ p∈P
p s
0
Except for finitely many primes (the prime P

factors of q), χ0 (p) = 1. By
Theorem 1.9 we can conclude that lims→1+ p∈P χ0 (p)p−s = ∞. Thus,
to prove Theorem 7.21, it is enough to show that the second term in
(7.17) is bounded as s → 1+.
Definition 7.23. Let χ be a Dirichlet character. Define the Dirich-
let L-function in Ω1 by
∞
X χ(n)
L(s, χ) := .
n=1
n−s
Since χ is multiplicative, the same argument that proved the Euler

product formula (Theorem 1.5) shows that
∞ Y
X χ(n) 1
= .
n=1
n−s p∈P
1 − χ(p)p−s
Note that
X
log L(s, χ) = − log(1 − χ(p)p−s )
p∈P
X χ(p)
−2s
= − − −s + O(p )
p∈P
p
X χ(p)
= + O(1).
p∈P
p−s
Therefore, to prove Theorem 7.21, it is enough to show that
lims→1+ L(s, χ) is finite and non-zero, for every non-trivial Dirichlet
character χ. If q = q1r1 . . . qkrk , then
Y 1
L(s, χ0 ) = −s
= (1 − q1−s ) . . . (1 − qk−s )ζ(s).
p∈P
1 − χ0 (p)p
Theorem 7.24. If χ is a non-trivial Dirichlet character, then

σc (L(s, χ)) = 0.
P∞
Proof: Since n=1 χ(n) does not converge, Theorem 3.12 yields
log |sN |
σc = lim supN →∞ log N ≥ 0. We can compute
q q
X X
χ(n) = χ(n)χ0 (n)
n=1 n=1
= φ(q) hχ, χ0 i`2 (Z∗q )
= 0.
Hence, by periodicity of χ, we can conclude that |sN | ≤ φ(q), and so
σc = 0.
Thus, lims→1+ L(s, χ) is finite, in fact, L(1, χ) is defined for every
non-trivial Dirichlet character χ. Hence, to prove Theorem 7.21, it
remains to show that L(1, χ) 6= 0, for χ 6= χ0 .
We will now fix a non-trivial character η ∈ Xq . We distinguish two
cases.
Case I: η is not a real-valued character.
Q
Lemma 7.25. For s > 1, χ∈Xq L(s, χ) ≥ 1.
Proof: By definition of the Dirichlet L-function and the power series
expansion of the natural logarithm, we have
!
Y Y X 1
L(s, χ) = exp log
χ∈X χ p∈P
1 − χ(p)p−s
q
90 7. CHARACTERS
∞
!
XXX χ(pk )
= exp
χ
p∈P k=1
kpks
∞
!
XX 1 X
= exp χ(pk ) , (7.18)
p∈P k=1
kpks χ
where rearranging the order of summation is justified by absolute con-

vergence. For any character χ, χ(1) = 1 and hence
1 X
hδ1 , χi = δ1 (m)χ(m)
φ(q) m∈Z∗
q
1
= .
φ(q)
Consequently, 1 = φ(q)hδ1 , χi, so that for any n ∈ N+ , we obtain
X X
χ(n) = φ(q) hδ1 , χiχ(n)
χ χ
= φ(q)δ1 (n)
≥ 0.
Q
We conclude by (7.18) that χ L(s, χ) = exp(r), where r is non-
negative.
Suppose that L(1, η) = 0. Then L(s, η) = O(s − 1) as s tends
to 1. Its conjugate η is also a character (and different from η, since
we assumed η takes on some non-real value somewhere). Moreover,
L(1, η) = ∞ η(n)
P
n=1 n = L(1, η) = 0. Thus, as s → 1+,
Y Y
L(s, χ) = L(s, χ0 ) · L(s, η) · L(s, η) · L(s, χ)
χ χ6=η,η,χ0
= O((s − 1)−1 ) O(s − 1) O(s − 1) O(1)
= O(s − 1),
which contradicts Lemma 7.25. This concludes case I.
Case II: η is real character.
Lemma 7.26. Let m ∈ N+ . Then n|m η(n) ≥ 0. If m = l2 with
P
l ∈ N+ , then n|m η(n) ≥ 1.

P
Proof: Write m = pr11 . . . prkk , then

k
r
X Y
η(n) = η(1) + η(pj ) + · · · + η(pj j ) .
n|m j=1
Since η is real, the only possible values for η(pj ) are 1, 0, and −1.
Corresponding to these cases, we observe that


rj + 1, if η(pj ) = 1,

1, if η(p ) = 0,
r j
η(1) + η(pj ) + · · · + η(pj j ) =
1, if η(pj ) = −1, and rj is even,


0, if η(pj ) = −1, and rj is odd.

P
Thus n|m η(n) is a product of non-negative factors. If m is a square,
P
all rj ’s are even, so that n|m η(n) is a product of numbers larger than
1.
Lemma 7.27. For all M ≤ N ∈ N+ and every σ > 0,

N
X η(n)
σ
= O(M −σ ).
n=M
n
Proof: Let sn = nk=1 η(k) and use summation by parts as follows

P
−1
N η(n)
X N
X
sn n − (n + 1) + O(M −σ )
−σ −σ

=
n σ
n=M n=M
N
X −1
−σ
n − (n + 1)−σ + O(M −σ )

≤ φ(q)
n=M
φ(q) M −σ − N −σ + O(M −σ )

=
= O(M −σ ),
where the estimate |sn | ≤ φ(q) was demonstrated in the proof of The-
orem 7.24.
For N ∈ N+ , set
X η(n)
SN = √ .
m,n≥1; mn≤N
mn
The following two claims imply that L(1, η) 6= 0 and thus conclude the
proof of Theorem 7.21.
Claim 1: SN ≥ c log N , for some c > 0.
Proof: Write
√
N X N b Nc
X η(n) X
−1/2
X X 1
SN = √ = k η(n) ≥ ,
k=1 mn=k
mn k=1 l=1
l
n|k
92 7. CHARACTERS
since Lemma 7.26 implies that if k = l2 , then n|k η(n) ≥ 1 and in

P
general, the sum is non-negative. But we √can estimate the last sum
R N
from below by comparing to the integral 1 dx x
≈ 12 log N .
√
Claim 2: SN = 2 N L(1, η) + O(1).
Before we prove this claim, we need the following approximation.
Lemma 7.28. For K ≥ 1, we have
K Z K+1
X 1 dx 1
√ = √ + τ + O( √ ),
m=1
m 1 x K
where τ is some positive constant.
R m+1
Proof: Let τm = √1m − m √dxx . Then
1 1
0 < τm < √ − √ , (7.29)
m m+1
which is an alternating series. So ∞
P
m=1 τm converges to some number
τ between 0 and 1. We have
K Z K+1 K
X 1 dx X
√ = √ + τm
m=1
m 1 x m=1
Z K+1 ∞
dx X
= √ +τ − τm .
1 x m=K+1
P∞
and by (7.29) we know that m=K+1 τm = O( √1K ).
We can now prove Claim 2.
Proof: (of Claim 2) Write
X 1 X η(n) X η(n) X 1
SN = √ √ + √ √
√ √ m nm≤N n √ n nm≤N m
m< N ,n> N n≤ N
= SI + SII .
The first term is easy to estimate using lemmata 7.27 and 7.28:
X 1 X η(n) X 1
SI = √ √ = √ O(N −1/4 ) = O(1).
√ m√ n √ m
m< N N <n≤N/m m< N
As for the second term, we have

X η(n) X 1
SII = √ √
√ n √ m
n≤ N m≤ N
7.5. NOTES 93
" r #
X η(n) Z Nn +1 dx n
= √ √ +τ +O
√ n 1 x N
n≤ N
" r ! r #
X η(n) N n
= √ 2 +1−1 +τ +O
√ n n N
n≤ N
" r r #
X η(n) N n
= √ 2 + (τ − 2) + O
√ n n N
n≤ N
√ X η(n) X η(n) X
η(n)O N −1/2 .

= 2 N + (τ − 2) √ +
√ n √ n √
n≤ N n≤ N n≤ N
The second term on the last line is√O(1) by Lemma 7.27, and the third
term is O(1) since there are only N terms in the sum. The first term
is a truncation of the series for L(1, η), so we get
∞
√ X η(n)
SII = 2 N [L(1, η) − √ ] + O(1),
√ n
n= N +1
which equals √
2 N L(1, η) + O(1)
by another application of Lemma 7.27.
We have thus proved Theorem 7.21.
7.4. Exercises
1. Let q be in N+ . Prove that gcd(n, q) = 1 if and only if there
exists m ∈ N+ such that mn ≡ 1, mod q.
2. Prove Proposition 7.22. (Hint: It is easy if G is cyclic. Then
show that G\1 × G2 = Ĝ1 × Ĝ2 ).
7.5. Notes
Theorem 7.15 is from [Hel69]. Our proof of Dirichlet’s theorem is
from [SS03]. This theorem was where Dirichlet series were first used
(and, in honor of this, were named after Dirichlet).
CHAPTER 8
Zero Sets
There is an interplay between the number of zeroes of a holomorphic

function and its size. Roughly speaking, the more zeroes a function
has, the larger it must be. The simplest example are polynomials —
if a polynomial has n zeroes, it must be of degree at least n thus
|P (z)| ≥ C|z|n as |z| → ∞. More generally, assume that f ∈ Hol (D)
is normalized so that f (0) = 1. Then, log |f (z)| is subharmonic in D
and so Z
0 = log |f (0)| ≤ log |f (z)| dA(z).
D
Thus log |f (z)| has to be “large enough” to offset the negativity of
log |f (z)| around points where f vanishes.
Definition 8.1. Let F be a family of holomorphic functions de-
fined on a set U and Z ⊂ U . We say that Z is a zero set for F, if there
exists a function f ∈ F that vanishes exactly on Z, that is, such that
Z = f −1 {0}.
It is well-known that for any connected open set U ⊂ C, the zero
sets for F = Hol (U ) are the sets Z ⊂ U that have no accumulation
points inside U .
For the Hardy spaces on the unit disk, the zero sets are well under-
stood:
Theorem 8.2. Let {λn }n ⊂ D be a sequence, 0 < p ≤ ∞. The
following are equivalent
• {λn } is zero set for H p (D),
{λn } is zero set for H ∞ (D),
• P
• Q n (1 − |λn |) < ∞,
• n |λλnn | 1−λ
z−λn
z
converges to a non-zero function.
n
The fact that the zero sets for H p (D) are independent of p follows
from inner-outer factorization. An analogous factorization theorem
does not hold for the polydisk and the zero sets for H p (Dn ) depend on
p when n > 1.
Precise descriptions of the zero sets for the Bergman space or the
Dirichlet space are not known.
95
96 8. ZERO SETS
Consider a Dirichlet series of the form f ∼ ∞ −ns

P
n=1 a2n 2 . Clearly,
2πi
if λ ∈ σc (f ) is a zero of f , then so is λ + log 2 k, for any k ∈ Z. The fol-
lowing theorem shows that the zero sets of Dirichlet series have similar
behavior at least in the half-plane Ωσu (f ) .
Theorem 8.3. Let f ∼ ∞ −s
P
n=1 an n , f (s0 ) = 0 and s0 > σu (f ).
Then, for every δ > 0, the strip {|Re (s − s0 )| < δ} contains infinitely
many zeroes.
Proof: Since the set of zeroes is discrete, we can find 0 < τ <
min {δ, σ0 − σu } such that C = ∂B(s0 , τ ) does not contain any zero
of f . By compactness, m := inf s∈C |f (s)| > 0. As the series converges
uniformly in Ωs0 −τ , we can find N ∈ N such that
N

f (s) −
X
−s
m
a n n < 4, for all s ∈ Ωs0 −τ
n=1
By Theorem 6.14, we can find an arbitrarily large t0 ∈ R so that for

all primes p ≤ N , t0 log p ≈ 0 mod 1. More precisely,
−σ it log n −σ m
n e 0 −n < , for all 1 ≤ n ≤ N, σ ∈ [σ0 −τ, σ0 +τ ].
4N (|an | + 1)
Consequently, by triangle inequality,
N
m X −s 3m
an n − n−s+it0 ≤

|f (s) − f (s + it0 )| ≤ + .
2 n=1
4
By Rouché’s theorem, it follows that f (s + it0 ) has a zero inside C,

that is, f (s) has a zero inside of C + it0 . Since t0 is arbitrarily large,
we can find infinitely many disjoint disks of this form.
We have an immediate corollary.
Corollary 8.4. If ϕ ∈ Mult (H2 ) and ϕ(s0 ) = 0 for some s0 ∈ Ω0 ,
then ϕ vanishes at infinitely many points.
Question 8.5. Does the above theorem hold for σc (f ) < σ0 <
σu (f )?
1
Note that the function ζ(s) has a zero s0 = 1 and no other zero in
the set {Re s > 1/2}, if the Riemann hypothesis holds, so the answer
to the question should be negative. In [MV, Problem 24], M. Balazard
poses the similar question of whether a convergent Dirichlet series can
have a single zero in a half-plane.
8. ZERO SETS 97
Let us define the uniformly local H p space on Ω1/2 by

Z θ+1 p1
p p
H∞ (Ω1/2 ) := g ∈ Hol (Ω1/2 ) : sup sup |g(σ+it)| dt < ∞ .
θ∈R σ>1/2 θ
Then, clearly,
H p (Ω1/2 ) ⊂ H∞
p
(Ω1/2 ),
and
p f (s)
f ∈ H∞ (Ω1/2 ) =⇒ ∈ H p (Ω1/2 ), for p > 1.
s
A deeper result is the following, which is a variant of Hilbert’s
inequality. See [Mon94] or [HLS97] for a proof.
Theorem 8.6. H2 ,→ H∞
2
(Ω1/2 ).
We define Hp by to be the completion of the set of all finite Dirichlet
series with respect to the norm
Z T 1/p
1 p
kf kHp := lim |f (it)| dt .
T →∞ 2T −T
Corollary 8.7. H2n ,→ H∞

2n
(Ω1/2 ), for all n ∈ N+ .
Question 8.8. Does Hp ,→ H∞
p
(Ω1/2 ) hold for all p > 1 (p ≥ 1)?
One might expect that the answer to the question has to be affir-
mative. As a warning, we recall a conjecture of Hardy and Littlewood
in 1935 that for any q ≥ 2 there exist a constant cq > 0 such that
Z 2π X q Z 2π X q

int

int

a n e dt ≤ c q

|a n |e dt.
0 0
The conjecture turns out to be true precisely when q is an even integer
(this was shown by Bachelis in 1973 [Bac73]).
Suppose that f ∈ H2 . Then f (s)
s
∈ H 2 (Ω1/2 ), and hence its zeroes
sk = σk + itk satisfy
X σk − 1/2
2
< ∞.
k
1 + |s k|
Also, if we define
X
A(θ) := σk − 1/2 ,
θ<tk <θ+1
then, by the above condition, A(θ) < ∞, for all θ ∈ R.

Theorem 8.9. (Hedelmalm, Lindqvist, Seip) If f ∈ H2 , and
f 6≡ 0, then supθ∈R A(θ) < ∞.
98 8. ZERO SETS
Proof: Suppose not, then there exist a sequence {θj }j ⊂ R such

that A(θj ) → ∞. Define fj (s) := f (s + θj ); then
kfj kH2 = kf kH2 .
Thus {fj }j is a bounded sequence in H2 , hence {fj (s)/s}j i is also
bounded in H 2 (Ω1/2 ). Let {sjk }k be the zeroes of fj (s)/s. The condition
A(θj ) → ∞ implies that
X σ j − 1/2
k
j 2 → ∞ as j → ∞. (8.10)
k
1 + |s k |
Using inner-outer factorization, this implies that fj ’s converge to 0
uniformly on compact sets, since the Blaschke product part does by
(8.10), and the outer parts are uniformly bounded on compact sets, by
the norm control. But by Proposition 7.11, some subsequence of {fj }j
converges
P uniformly on compact subsets to a vertical limit function
fχ = n an χ(n)n−s , where χ is a character. We conclude that fχ ≡ 0,
a contradicton.
Question 8.11. How can the zero sets of H2 , Hw2 , etc. be classified?
CHAPTER 9
Interpolating Sequences
9.1. Interpolating Sequences for Multiplier algebras

Definition 9.1. Let H be a Hilbert space of analytic functions on
X with reproducing kernel k. We say that (λn ) ⊂ X is an interpolating
sequence for Mult (H), if
{(ϕ(λn )), ϕ ∈ Mult (H)} = `∞ .
In other words, we require that the map E : φ 7→ (φ(zn )) maps

Mult (H) onto `∞ (it always maps into, by Proposition 11.9). By ba-
sic functional analysis, whenever one has an interpolating sequence, it
comes with an interpolation constant.
Indeed, consider the quotient Banach space X = Mult (H)/N ,
where
N := {f : f (zn ) = 0, ∀ n ∈ N}
is the kernel of E. We obtain a bounded operator Ẽ : X → `∞ , which is
one-to-one and onto. By the open mapping theorem, it has a bounded
inverse. We conclude that if {zn }n is an interpolating sequence for
Mult (H), then there exists a constant C > 0 such that for any sequence
(an )n ∈ `∞ , there exists a function f ∈ Mult (H) such that f (zn ) = an
for all n ∈ N and kf k∞ ≤ Ck(an )k∞ . The infimum of those C for
which this holds is called the interpolation constant of the sequence.
The exact description of interpolating sequences for particular
spaces is hard. There is a general result due to S. Axler [Axl92] show-
ing that sequences that tend to the boundary will, in many spaces,
have subsequences that are interpolating, but verifying the condition
of the theorem can be difficult.
Theorem 9.2. (Axler) Let H be a separable reproducing kernel

Hilbert space on X, and assume that Mult (H) separates points of X.
Suppose that (xn ) is a sequence with the property that for any subse-
quence (xnk ), there exists some φ ∈ Mult (H) such that limk→∞ φ(xnk )
does not exist. Then (xn ) has a subsequence that is an interpolating
sequence for Mult (H).
99
100 9. INTERPOLATING SEQUENCES
L. Carleson in 1958 characterized interpolating sequences for

H ∞ (D). For later convenience, we shall apply a Cayley transform and
quote the result for H ∞ (Ω0 ). First we need to introduce a metric.
Definition 9.3. Let A be a normed algebra of functions on the
set X. We define the Gleason distance ρA between two points x and y
by
ρA (x, y) = sup{|φ(y)k : φ(x) = 0, kφk ≤ 1}.
When the algebra is understood, we shall write ρ.
For the algebra H ∞ (D), the Gleason distance is called the pseudo-
hyperbolic metric, and , and it is given by

z−w
ρH ∞ (D) (z, w) = .
1 − w̄z
In the right half-plane, this becomes

s − u
ρH ∞ (Ω0 ) (s, u) = .
s + ū
In the polydisk, it is straightforward to show

zj − wj
ρH ∞ (Dm ) (z, w) = max . (9.4)
1≤j≤m 1 − w̄j zj
Theorem 9.5. (Carleson) Let (sj ) ⊂ Ω0 . Then the following are

equivalent:
(1) (sj ) is an interpolating sequence for H ∞ (Ω0 ).
Q si −sj
(2) inf j i6=j si +s̄j > 0.

si −sj
(3) inf i6=j si +s̄j > 0 and there exists C > 0 such that for every
f ∈ H 2 (Ω0 ),
X
σj |f (sj )|2 ≤ Ckf k22 .
j
Carleson’s theorem is very important, and the various conditions in

it have names.
Definition 9.6. Let A be a normed algebra of functions on the
set X, and let ρ = ρA be the Gleason distance. We say a sequence (xn )
is weakly separated if inf m6=n ρ(xm , xn ) > 0.
We say the sequence is strongly separated if
inf [sup{|φ(xn )| : φ(xm ) = 0 ∀ m 6= n, kφk ≤ 1}] > 0. (9.7)
n
9.1. INTERPOLATING SEQUENCES FOR MULTIPLIER ALGEBRAS 101
In H ∞ (Ω0 ), a sequence is strongly separated if and only if the a

priori stronger condition
" #
Y
inf ρ(sm , sn ) > 0
n
m6=n
holds; this is an elementary consequence of the fact that dividing out

by a Blaschke product does not increase the norm. In the polydisk, as
we shall see in Theorem 9.11, these two conditions are different.
Definition 9.8. Let H be a reproducing kernel Hilbert space on
X, and let µ be a measure on X. We say µ is a Carleson measure for
H if there exists a constant C such that
Z
|f |2 dµ ≤ Ckf k2H ∀ f ∈ H.
With these definitions, condition (2) in Carleson’s theorem becomes

the statement that the sequence is strongly separated, and condition
P
(3) is that the sequence is weakly separated and the measure σj δsj
2
is a Carleson measure for H (Ω0 ).
Question 9.9. What are the interpolating sequences for
Mult (Hw2 )?
The answer is not known in general, but K. Seip [Sei09] showed
that for bounded sequences, the interpolating sequences for Mult (H2 )
are the same as for the much larger space H ∞ (Ω0 ). Let us use H∞ to
denote Mult (H2 ), which by Theorem 6.42 is the bounded functions in
Ω0 that have a Dirichlet series:
H∞ = H ∞ (Ω0 ) ∩ D.
∞
We shall write Hm for those f in H∞ whose Dirichlet series is supported
on Nm .
Theorem 9.10. (Seip) Let (sj ) be a bounded sequence in Ω0 . Then
the following are equivalent:
(i) It is an interpolating sequence for H∞ .
(ii) It is an interpolating sequence for H2∞ .
(iii) It is an interpolating sequence for H ∞ (Ω0 ).
Moreover, if {sj } is contained in a vertical strip of height less than
2π
log 2
, and is bounded horizontally, these three conditions are equivalent
to (sj ) being an interpolating sequence for H1∞ .
To prove Seip’s theorem, we need a result by B. Berndtsson, S.-Y.
Chang and K.-C. Lin [BCL87] that gives a sufficient condition for a
sequence to be interpolating on the polydisk. We shall explain what

condition (3) means in Section 9.2.
Theorem 9.11. (Berndtsson, Chang and Lin) Consider the
three statements
(1) There exists c > 0 such that
Y
ρH ∞ (Dm ) (λi , λj ) ≥ c
j6=i
for all i.
(2) The sequence {λi }∞ ∞ m
i=1 is an interpolating sequence for H (D ).
(3) The sequence {λi }∞ i=1 is weakly separated and the associated
Grammian with respect to Lebesgue measure σ is bounded.
Then (1) implies (2) and (2) implies (3). Moreover the converse of
both these implications is false.
To prove Seip’s theorem, we need to compare Gleason differences
in different algebras. For the remainder of the section, we shall adopt
the following notation:

zj − wj
dm (z, w) = ρH ∞ (Dm ) (z, w) = max
1≤j≤m 1 − w̄j zj

s − u
ρ(s, u) = ρH ∞ (Ω0 ) (s, u) =
s + ū
ρm (s, u) = dm ((2−s , ., p−s
m ), (2
−u
, ., p−u
m ))
For points s, u in Ω0 , we shall write

s = σ + it, u = υ + iy.
Lemma 9.12. For each n ≥ 2,
d1 (n−s , n−u ) ≤ ρ(s, u)
ρ2 (s, u) ≤ ρ(s, u).
Proof: The first inequality is because the map s 7→ n−s is a holo-
morphic map from Ω0 to D, so it is tautologically distance decreasing
in the Gleason distances for the corresponding H ∞ spaces.
The second inequality follows from the first. 2
Lemma 9.13. For every M > 0, there exists γ > 0 such that if
s, u ∈ Ω0 and |s|, |u| ≤ M , then
ρ2 (s, u) ≥ ρ(s, u)γ .
2π
If in addition |t − y| ≤ H < log 2
, then we can choose γ so that
ρ1 (s, u) ≥ ρ(s, u)γ .
9.1. INTERPOLATING SEQUENCES FOR MULTIPLIER ALGEBRAS 103
Figure 1. Curve ρ2 = ργ fits outside shaded area
Proof: It is sufficient to prove that there exist constants c, C > 0

such that
(1) ρ2 (s, u) ≥ cρ(s, u)
(2) 1 − ρ2 (s, u) ≤ C(1 − ρ(s, u)).
(See Figure 9.1).
To prove (1), let K be the closed semi-disk
K = {s : σ ≥ 0, |s| ≤ M }.
Define the function ψ on K × K by
 ρ(s,u)
 ρ2 (s,u) if s 6= u, and s, u ∈ Ω0

ψ = 1 if <s or <u = 0
 2σ −2−σ if s = u ∈ Ω .

2σ log 2 0
It is straightforward to check that ψ is continuous, so we can set

c = 1/ max ψ(s, u)
K×K
and get (1).

For (2), we observe that
−s
p − p−u

ρ2 (s, u) = max .
p=2,3 1 − p−s−ū
Writing s = σ + it and u = υ + iy, we get

s − u 2

2
1 − ρ(s, u) = 1 −
s + ū
4συ
= . (9.14)
(σ + υ)2 + (t − y)2
We also have
p−2σ + p−2υ − 2<p−s−ū

2
1 − ρ2 (s, u) = min 1 −
p=2,3 1 − 2<p−s−ū + p−2σ−2υ
1 + p−2σ−2υ − p−2σ − p−2υ
= min
p=2,3 1 − 2<p−s−ū + p−2σ−2υ
(1 − p−2σ )(1 − p−2υ )
= min (9.15).
p=2,3 (1 − p−σ−υ )2 + 2p−σ−υ (1 − cos[log p(t − y)])
We would like to show that for some constant CM we have that

συ
1 − ρ2 (s, u)2 ≤ CM , (9.16)
(σ + υ) + (t − y)2
2
as this, together with (9.14), would give (2).

First, assume that
2π
|t − y| ≤ H < . (9.17)
log p
Then, as 1−p−x is comparable to x on [0, 2M ], we see that the numera-
tor in (9.15) is comparable to συ, and the first term in the denominator
is comparable to (σ + υ)2 . As for the second term, a Taylor series ar-
gument shows that for t − y close to 0,
1 − cos[log p(t − y)] ≈ (t − y)2 . (9.18)
Continuity and compactness show that (9.18) remains true (with some
constants) if (9.17) holds, as the left-hand side can then vanish only at
t − y = 0. This gives us the second part of the lemma, where we only
need to use the prime p = 2.
If (9.17) fails with p = 2, there will be points where t 6= y but
1 − cos[log 2(t − y)] = 0.
However, one cannot simultaneously have
1 − cos[log 3(t − y)] = 0,
9.2. INTERPOLATING SEQUENCES IN HILBERT SPACES 105
since log 2 and log 3 are rationally linearly independent. So by com-

pactness and continuity again, we get (9.16).
Proof of Thm. 9.10: Suppose (sj ) is bounded and interpolating
for H ∞ (Ω0 ). Then by Theorem 9.5 and Lemma 9.13, we have
Y
inf ρ2 (si , sj ) ≥ 0.
j
i6=j
Therefore by Theorem 9.11, the sequence ((2−sj , 3−sj )) is interpolating

for H ∞ (D2 ). So if (aj ) is any target in `∞ , there exists some ψ ∈
H ∞ (D2 ) satisfying
ψ(2−sj , 3−sj ) = aj .
Then
φ(s) = ψ(2−s , 3−s )
solves the interpolation problem in H ∞ (Ω0 ) ∩ D.
Finally, if the vertical height of a rectangle containing all the points
is less than 2π/ log 2, the second part of Lemma 9.13 shows that one
can interpolate with a function of the form ψ(2−s ), where ψ ∈ H ∞ (D).
9.2. Interpolating sequences in Hilbert spaces

Let Hk be a reproducing kernel Hilbert space on a set X. Given a
sequence (λi ) in X, let gi denote the normalized kernel function at λi :
1
gi := kλ .
kkλi k i
Define a linear operator E by
E : f 7→ hf, gi i. (9.19)
We say the sequence (λi ) is an interpolating sequence for Hk if the map
E is into and onto `2 . (Note that because gi is normalized, E necessarily
maps into `∞ ; but it does not have to map into `2 ).
We say that a set of vectors {vi } in a Banach space is topologically
free if no one is contained in the closed linear span of the others. This
is equivalent to the existence of a dual system, vectors {hi } in the dual
satisfying
hhj , vi i = δij .
The dual system is called minimal if each hj is in ∨{vi }.
Theorem 9.20. The sequence (λi ) is an interpolating sequence for
Hk if and only if the Gram matrix G = hgj , gi i is bounded and bounded
below.
Proof: (⇒) Suppose E is bounded and onto `2 . As E ∗ ej = gj , we

have
hgj , gi i = hEE ∗ ej , ei i
is bounded. By the open mapping theorem, E has an inverse
E −1 : `2 → ∨{gi } ⊆ Hk .
Let hj = E −1 ej . Then
G−1 = hhj , hi i
is bounded.
(⇐) Suppose G is bounded and bounded below. Define
L : `2 → Hk
ej 7 → gj .
Since G is bounded, L is bounded, and E = L∗ is therefore a bounded
map into `2 . Since G is bounded below, the minimal dual system {hj }
to {gj } has a bounded Gram matrix (see Exercise 9.37), and if (aj ) is
any sequence in `2 , we have
X
E( aj hj ) = (aj ),
so E is onto.
Theorem 9.21. Any interpolating sequence for Mult (Hk ) is an

interpolating sequence for Hk .
Proof: Suppose (λi ) is an interpolating sequence for Mult (Hk ).
Then there is a constant M such that for every sequence (wi ) in the
unit ball of `∞ , the map
R : gi 7→ w̄i gi
extends to a linear operator on Hk of norm at most M (since it is
the adjoint of a multiplication operator that solves the interpolation
problem). Therefore, for all finite sequences of scalars (cj ), we have
X X
k cj w̄j gj k2 ≤ M 2 k cj g j k 2 .
Write this as
X X
cj c̄i w̄j wi hgj , gi i ≤ M 2 cj c̄i hgj , gi i,
i,j i,j
let wj = e2πitj and integrate with respect to each tj to get

X X
|cj |2 ≤ M 2 cj c̄i hgj , gi i.
i,j
9.3. THE PICK PROPERTY 107
This proves G is bounded below. A similar argument, with wj = e2πitj

and cj = aj e2πitj gives
X X
aj āi hgj , gi i ≤ M 2 |aj |2 ,
i,j
so G is also bounded. By Theorem 9.20, we are done.

Interpolating sequences for H2 and Hw2 where the weights wn are
(log n)α , as in (6.34), are studied in [OS08]. In particular, they show
that for bounded sequences, the interpolating sequences are the same
as in the corresponding space of analytic functions that do not have to
have Dirichlet series representations.
9.3. The Pick property

A particularly useful feature of the Hardy space H 2 is that it has
the Pick property.
Definition 9.22. The reproducing kernel Hilbert space Hk on X
has the Pick property if, for every subset F ⊆ X, and every function
ψ : F → C, if the linear operator defined by
T : kλ 7→ ψ(λ)kλ
is bounded by C on ∨{kλ : λ ∈ F }, then there is a multiplier φ of Hk ,
with multiplier norm bounded by C, and satisfying
φ(λ) = ψ(λ) ∀ λ ∈ F.
Theorem 9.23. If Hk has the Pick property, then the interpolating
sequences for Mult (Hk ) and Hk coincide.
Proof: Suppose (λi ) is an interpolating sequence for Hk , so there
are constants c1 and c2 so that
X X X
c1 |ai |2 ≤ k ai gi k2 ≤ c2 |ai |2 .
Let (wi ) be a sequence in the unit ball of `∞ . Define R by
R : gi 7→ w̄i gi .
Then

c2 ∗ c2
h − R R gj , gi i = hgj , gi i − wi w̄j hgj , gi i
c1 c1
c2
≥ (c1 δij ) − wi w̄j (c2 δij )
c1
= c2 δij (1 − |wi |2 )
≥ 0.
p
Therefore R is bounded by c2 /c1 , so byp
the Pick property, there is a
multiplier φ of Hk with norm bounded by c2 /c1 such that φ(λi ) = wi .

The idea of using the Pick property to reduce the characterization
of interpolating sequences for a multiplier algebra to the more tractable
problem of characterizing them for a Hilbert space was originally due
to H.S. Shapiro and A. Shields, in the case of H ∞ (D) [SS61]. It was
developed more systematically by D. Marshall and C. Sundberg in
[MS94].
The space H2 does not have the Pick property — one way to see
this is that the bounded interpolating sequences for H2 are interpo-
lating sequences for H 2 (Ω1/2 ) [OS08], whereas bounded interpolating
sequences for the multiplier algebra can only accumulate on the bound-
ary of Ω0 by Theorem 9.10. However, there are several Hilbert spaces
of Dirichlet series that have the Pick property (and a stronger, matrix-
valued version, called the complete Pick property).
Theorem 9.24. If k(s, u) = η(s + ū), then this has the complete
Pick property for each of the following η’s:
1
η(s) = (9.25)
2 − ζ(s)
ζ(s)
η(s) =
ζ(s) + ζ 0 (s)
ζ(2s)
η(s) =
2ζ(2s) − ζ(s)
P (2)
η(s) = . (9.26)
P (2) − P (2 + s)
In (9.26), the function P (s) is the prime zeta function , defined by
X
P (s) = p−s .
p∈P
Definition 9.27. A sequence (λi ) satisfies Carleson’s condition in

the reproducing kernel Hilbert space Hk if there exists a constant C so
that
X |f (λi )|2
2
≤ Ckf k2 ∀ f ∈ Hk .
i
kk λ i
k
Definition 9.28. The sequence (λi ) is weakly separated in the
reproducing kernel Hilbert space Hk if there exists a constant c > 0 so
that, for all i 6= j, the normalized reproducing kernels satisfy
|hgi , gj i| ≤ 1 − c.
9.4. SAMPLING SEQUENCES 109
Theorem 9.29. [AHMR17] Let Hk have the complete Pick prop-

erty. Then a sequence (λi ) is an interpolating sequence if and only if
it is weakly separated and satisfies Carleson’s condition.
9.4. Sampling sequences
Definition 9.30. Let K be a Hilbert space of functions on a set
X with bounded point evaluations and denote the reproducing kernel
at ζ ∈ X by kζ . We say that a sequence {zn }n ⊂ X is a sampling
sequence, if for all f ∈ K, we have
X |f (zn )|2
2
≈ kf k2K .
n
kkzn k
Equivalently,one can say that the operator E : K → `2 given by
f (zn )
E : f 7→ kk zn k n
is bounded and bounded below. Another way to
rephrase this is to require that the sequence of normalized reproducing
kzn
kernels kkz k n forms a frame.
n
For weighted Bergman spaces on the disk there is a complete de-
scription of sampling sequences in terms of lower density of the se-
quence.
Proposition 9.31. There are no sampling sequences for the Hardy
space of the disk.
Proof: Suppose that {zn }n ⊂ D is a sampling sequence for H 2 (D).
Then {zn }n cannot be a Blachke sequence, since the corresponding
Blaschke product f would satisfy 0 < kf k2 < ∞ and n |f (zn )|2 ·
P
kkzn k−2 = 0. If {zn }n is not a Blaschke sequence, we consider the
function f (z) = 1. Then
X |f (zn )|2 X
2
= (1 − |zn |2 ) = ∞,
n
kk z n k n
while kf k2 < ∞, a contradiction.

However, there exists “generalized sampling sequences” for the
Hardy space, that is, sequences satisfying
X |f 0 (zn )|2
|f (0)|2 + ≈ kf k22 ,
k k̃ k2
n zn
where k̃zn is the reproducing kernel for the derivative. But this is
because differentiation maps the Hardy space (modulo constants) iso-
metrically onto a weighted Bergman space.
There is another proof of the fact that H 2 (D) does not admit any
sampling sequence, using the fact that multiplication by z is isometric.
The same idea works for H2 . The reader is invited to recast that proof
for the Hardy space.
Proposition 9.32. There are no sampling sequences on H2 .
Proof: The multiplication operator MN −s is an isometry on H2 .
On the other hand, the sequence fN := MN −s f tends to 0 uniformly
P N (sn )|2
on compact sets in Ω1/2 . Thus n |fkk sn k
2 → 0 as N → ∞ and thus
2 2
cannot be comparable to kfN k = kf k .
A similar argument shows that “generalized sampling sequences”
involving f 0 (sn ) do not exists.
Question 9.33. Is there a sensible interpretation of “{sn }n is a
generalized sampling sequence for H2 ”? If so, how are these character-
ized?
9.5. Exercises
Exercise 9.34. Prove Equation (9.4). (Hint: use an automorphism
of Dm to move one point to the origin).
Exercise 9.35. Prove that any sequence that tends sufficiently
quickly to ∂D is an interpolating sequence for H ∞ (D).
Exercise 9.36. Fill in the details of the proof of (9.16).
Exercise 9.37. Prove that if (hi ) is the minimal dual system of
(gi ), then the inverse of the Gram matrix G = hgj , gi i is the matrix
hhj , hi i.
9.6. Notes
For a much more comprehensive treatment of interpolating se-
quences, we recommend the excellent monograph [Sei04] by K. Seip.
For a concise treatment for H ∞ (D) and H 2 (D), including Theorem 9.20,
see [Nik85].
Axler’s theorem 9.2 was proved for multipliers of the Dirichlet space
[Axl92], but the argument readily adapts to the stated version. Car-
leson’s theorem is in [Car58]. Seip’s paper [Sei09] contains much
more information on interpolating sequences for H∞ than Theorem 9.10
alone.
Necessary and sufficient conditions for a sequence to be interpo-
lating for H ∞ (D2 ) are given in [AM01], but they do not completely
resolve the issue. For example, the following is still open:
Question 9.38. If λn is strongly separated in H ∞ (D2 ), is it an
interpolating sequence?
9.6. NOTES 111
Interpolating sequences for H2 and Hw2 were first considered in

[OS08]. See also [Ols11].
The fact that (9.25) gives rise to a Pick kernel was observed in
[McCa04]. Necessary and sufficient conditions for a general kernel to
have the complete Pick property, a matrix valued version of the Pick
property, are given by P. Quiggin [Qui93, Qui94] and S. McCullough
[McCu92, McCu94]; see also [AM00]. The application to kernels
of the form discussed in Theorem 9.24 is discussed in [?]. The kernel
coming from (9.26) is particularly interesting, as it is in some sense
universal amongst all kernels with the complete Pick property. See [?]
for details.
CHAPTER 10
Composition operators
Definition 10.1. Let K be a Hilbert space of analytic functions

on X with reproducing kernel k and let ϕ : X → X be an ana-
lytic function. To ϕ we associate a composition operator Cϕ given
by Cϕ (f ) := f ◦ ϕ.
The study of such operators was originally inspired by the following
result.
Theorem 10.2. (Littlewood’s subordination principle) For
any analytic ϕ : D → D, the operator Cϕ is bounded on H 2 (D).
J. Shapiro proved in 1987 [Sha87] that Cϕ is compact on H 2 (D),
if and only if “ϕ does not get too close to ∂D too often.”
An interesting property of composition operators is that their ad-
joints permute the kernel functions. Indeed,
hf, Cϕ∗ kζ i = hCϕ f, kζ i
= hf ◦ ϕ, kζ i
= f (ϕ(ζ))
= hf ◦ ϕ, kζ i
= hf, kϕ(ζ) i,
so Cϕ∗ kζ = kϕ(ζ) .
Recently, various properties of Cϕ were studied in terms of prop-
erties of ϕ on the Hardy space, the Dirichlet space and the Bergman
space.
We now gather some results about composition operators on H2 .
Let Φ : Ω1/2 → Ω1/2 be an analytic function. Note that CΦ : f 7→
f ◦ ΦPmight not map Dirichlet P series to Dirichlet series. Indeed, if
∞
f ∼ n=1 an n , then (f ◦ Φ) ∼ n an n−Φ(s) . The next two theorems
−s
are due to J. Gordon and H. Hedenmalm [GH99].

Theorem 10.3. An analytic function Φ : Ω1/2 → Ω1/2 gives rise to
a composition operator CΦ : H2 → D, if and only if Φ(s) = c0 s + ϕ(s),
where c0 ∈ N and ϕ ∈ D.
113
114 10. COMPOSITION OPERATORS
Theorem 10.4. (Gordon, Hedenmalm) An analytic function

Φ : Ω1/2 → Ω1/2 gives rise to a bounded composition operator CΦ :
H2 → H2 , if and only if Φ(s) = c0 s + ϕ(s), where c0 ∈ N, ϕ ∈ D, and
Φ has an analytic extension to Ω0 such that Φ(Ω0 ) ⊂ Ω0 , if c0 > 0 and
Φ(Ω0 ) ⊂ Ω1/2 , if c0 = 0.
They also proved that CΦ is a contraction (i.e., kCΦ k ≤ 1), if and
only if c0 > 0 in the above theorem. Furthermore, the same theorem
holds for Hp with 2 ≤ p < ∞ and the conditions are necessary for
1 < p < 2.
Compactness of composition operators was studied by F. Bayart.
He proved the following theorem [Bay03]:
Theorem 10.5. (Bayart) The composition operator CΦ is compact
on Mult (Hw2 ), if and only if Φ(Ω0 ) ⊂ Ωε , for some ε > 0.
He also proved that if CΦ is a compostion operator on H2 , then
QCΦ Q−1 is a composition operator on H 2 (T∞ ), i.e., there exists ψ :
D∞ ∩ `2 → D∞ ∩ `2 such that Cψ = QCΦ Q−1 . This allows one to
construct compact composition operators on H2 that are not Hilbert-
Schmidt.
CHAPTER 11
Appendix
11.1. Multi-index Notation

When dealing with power series in several variables, it is easy to
become overwhelmed with subscripts. Multi-index notation is a way
to make formulas easier to read.
We fix the number of variables, d say, and assume that is under-
stood. We write
α = (α1 , α2 , . . . , αd )
for a multi-index, where α is in Nd or Zd . Then
X
cα z α
stands for X
cα1 ,α2 ,...,αd z1α1 z2α2 · · · zdαd .
We define
d
X
|α| = |αr |
r=1
α! = α1 !α2 ! · · · αd !
11.2. Schwarz-Pick lemma on the polydisk

Schwarz’s lemma on the disk has a non-infinitesimal version, called
the Schwarz-Pick lemma. Both these lemmata generalize to the poly-
disk.
Lemma 11.1. (Schwarz-Pick) If f : D → D is holomorphic, then

f (w) − f (z) w−z
≤ ,

1 − wz

1 − f (w)f (z)
for all z, w ∈ D.
Proof: For ξ ∈ D, let ψξ be the automorphism of the disk that
ξ−z
exchanges 0 and ξ, that is, ψξ (z) = 1−ξz . Consider the function g :
115
116 11. APPENDIX
D → D given by g = ψf (w) ◦ f ◦ ψw . Choose ζ = ψw (z) so that

f (w) − f (z)
|g(ζ)| = |(ψf (w) ◦ f ◦ ψw )(ψw (z))| = |ψf (w) (f (z))| =

1 − f (w)f (z)
and
w−z
|ζ| = |ψw (z)| = .
1 − wz
Also, g(0) = 0 so that |g(ζ)| ≤ |ζ| by the classical Schwarz lemma.
Lemma 11.2. Schwarz’s lemma on the polydisk Let f ∈

H ∞ (DN ) satisfies kf k∞ ≤ 1 and f (0) = 0. Then
|f (w1 , . . . , wN )| ≤ max |wi |.
1≤i≤N
Proof: Let
r = max |wi |.
i=1,...,N
∞
Define g ∈ H (D) by
z
g(z) := f ( (w1 , . . . , wN )).
r
Then kgk∞ ≤ 1, and g(0) = 0. Apply Schwarz’s lemma to g to conclude
|g(r)| ≤ r.
Lemma 11.3. Let f ∈ H ∞ (D) satisfies kf k∞ ≤ K and f (0) = 1.

Then f 6= 0 on K1 D.
Proof: We may assume that g is non-constant. Consider g(z) =
f (z)
K
then g(0) = 1/K and g : D → D. If f (z) = 0, then, by the
,
Schwarz-Pick lemma applied to g and w = 0

1 1 f (0) − 0 g(0) − g(z)
0−z

K
= = ≤ = |z|.

K 1 − f (0)/K · 0 1 − g(0)g(w) 1 − 0 · z
1
Thus f cannot vanish on K
D.
Lemma 11.4. Let f ∈ H ∞ (DN ) satisfies kf k∞ ≤ K and f (0) = 1.

Then f 6= 0 on K1 DN .
Proof: Fix w = (w1 , . . . , wN ) ∈ DN , and define |w|∞ =
maxi=1,...,N |wi |. Define g ∈ H ∞ (D) by g(z) := f ( |w|
zw
∞
), then kgk∞ ≤
K. If f (w) = 0, then g(|w|∞ ) = 0. Thus, by the preceding lemma,
|w|∞ ≥ 1/K.
11.3. REPRODUCING KERNEL HILBERT SPACES 117
11.3. Reproducing kernel Hilbert spaces

Let H be a Hilbert space of functions on a set X such that eval-
uation at each point of X is continuous. (Note: when we speak of a
Hilbert space of functions on X, we assume that any function that is
identically zero on X is zero in the Hilbert space). Then by the Riesz
representation theorem, for each w ∈ X, there must be some function
kw ∈ H such that
f (w) = hf, kw i.
One can think of kw as a function in its own right, kw (z) say. We call
the function k(z, w) = kw (z) the kernel function for H, and we call kw
the reproducing kernel at w.
Proposition 11.5. Let H be a Hilbert function space on X, and
let {ei }i∈I be any orthonormal basis for H. Then
X
k(z, w) = ei (w)ei (z). (11.6)
i∈I
Proof: This is just Parseval’s identity:

k(z, w) = hkw , kz i
X
= hkw , ei i waei , kζ i
i∈I
X
= ei (w)ei (z).
i∈I
It follows from (11.6) that k(z, w) = k(w, z).

Proposition 11.7. Let H be a Hilbert space of analytic functions
on a topological space X such that the function κ : X → H given by
κ(w) := kw is continuous. Let {fn }n∈N ⊂ H be a bounded sequence.
Then, the following are equivalent
(1) hfn , gi → hf, gi for all g in some set S ⊂ H, whose span is
dense in H,
(2) fn → f weakly in H,
(3) fn → f uniformly on compact subsets of X,
(4) fn → f pointwise in X.
Proof: (1) =⇒ (2) : By linearity, hfn , gi → hf, gi for all g ∈
span S. Now choose an arbitrary g ∈ H, fix ε > 0 and find g0 ∈ span S
such that kg − g0 k < ε. Then
lim |hfn − f, gi| ≤ lim |hfn − f, g − g0 i| + lim |hfn − f, g0 i|
n→∞ n→∞ n→∞
≤ lim kfn − f k · kg − g0 k + 0
n→∞
118 11. APPENDIX
≤ M ε,
where M = supn∈N kfn k. Since ε was arbitrary, we conclude that fn →
f weakly.
(2) =⇒ (3) : Let K ⊂ X be compact, then by continuity of κ,
the set K̃ := {kw ; w ∈ K} is also compact. Fix ε > 0 and find a
finite ε-net {kw1 , . . . , kwm } in K̃. Find N ∈ N such that for all n > N
hfn − f, kwj i < ε holds for j = 1, . . . , m. Then for any w ∈ K and
n > N:
|fn (w) − f (w)| = |hfn − f, kw i|
≤ |hfn − f, kwi i| + |hfn − f, kw − kwi i|
≤ ε + kfn − f k · kkw − kwi k
≤ ε + 2M ε
= (2M + 1)ε,
for a suitable i (such i exists since {kw1 , . . . , kwm } is an ε-net). Since
ε > 0 was arbitrary, we conclude that fn → f uniformly in K.
(3) =⇒ (4) : Obvious.
(4) =⇒ (1) : Follow immediately, since (4) means that (1) holds
with S = {kw }w∈X
Corollary 11.8. Let {fn }n∈N be a bounded sequence with H as in

Proposition 11.7. Then there exists a subsequence that satisfies all the
equivalent conditions of Proposition 11.7.
Proof: Since any bounded set in a Hilbert space weakly sequen-
tially compact, there exists a subsequence that converges weakly. By
Proposition 11.7, it satisfies all four conditions.
11.4. Multiplier Algebras

If H is a Hilbert space of functions on X, we let Mult (H) denote
the multiplier algebra, i.e. the set
Mult (H) = {φ : φf ∈ H ∀ f ∈ H}.
It follows from the closed graph theorem that if φ is in Mult (H), then
the operator Mφ of multiplication by φ is bounded. The adjoint Mφ∗
has all the kernel functions as eigenvectors.
let φ be in Mult (H). Then
Mφ∗ kw = φ(w)kw , ∀ w ∈ X. (11.10)
kMφ k ≥ sup |φ|. (11.11)
X
11.4. MULTIPLIER ALGEBRAS 119
If the norm on H is an L2 -norm on X, then (11.11) becomes an equal-

ity.
Proof: Let f be an arbitrary function in H. Then
hf, Mφ∗ kw i = hφf, kw i
= φ(w)f (w)
= hf, φ(w)kw i.
This proves (11.10).
As
kMφ∗ k ≥ sup kMφ∗ kw k/kkw k
w∈X
= sup |φ(w)|,
w∈X
we get (11.11).
Finally, if the norm on H is the L2 (µ)-norm, then the inequality
Z Z
2 2
|φf | dµ ≤ kφk∞ |f |2 dµ
X X
means kMφ k ≤ kφk∞ .

assume Mult (H) separates the points of X. Then Mult (H) equals its
commutant in the bounded linear operators on H.
Proof: Suppose T is in the commutant of Mult (H). Then T ∗ has
each kernel function kw as an eigenvector, since Mult (H) separates the
points of X. Therefore
T ∗ kw = φ(w)kw ,
for some function φ. Therefore T = Mφ , and since T is bounded, this
means φ is a multipler. 2
Bibliography
[AHMR17] Alexandru Aleman, Michael Hartz, John E McCarthy, and Stefan

Richter, Interpolating sequences in spaces with the complete pick prop-
erty, International Mathematics Research Notices (2017), rnx237.
[AM00] J. Agler and J.E. McCarthy, Complete Nevanlinna-Pick kernels, J.
Funct. Anal. 175 (2000), no. 1, 111–124.
[AM01] , Interpolating sequences on the bidisk, International J. Math. 12
(2001), no. 9, 1103–1114.
[Axl92] S. Axler, Interpolation by multipliers of the Dirichlet space, Quart. J.
Math. Oxford Ser. 2 43 (1992), 409–419.
[Bac73] Gregory F. Bachelis, On the upper and lower majorant properties in
Lp (G), Quart. J. Math. Oxford Ser. (2) 24 (1973), 119–128.
[Bay03] Frédéric Bayart, Compact composition operators on a Hilbert space of
Dirichlet series, Illinois J. Math. 47 (2003), no. 3, 725–743.
[BCL87] B. Berndtsson, S.-Y. Chang, and K.-C. Lin, Interpolating sequences in
the polydisk, Trans. Amer. Math. Soc. 302 (1987), 161–169.
[Bes32] A.S. Besicovitch, Almost periodic functions, Cambridge University Press,
London, 1932.
[BH31] H. F. Bohnenblust and Einar Hille, On the absolute convergence of
Dirichlet series, Ann. of Math. (2) 32 (1931), no. 3, 600–622.
[Boa97] Harold P. Boas, The football player and the infinite series, Notices Amer.
Math. Soc. 44 (1997), no. 11, 1430–1435.
[Boh13a] H. Bohr, Über die Bedeutung der Potenzreihen unendlich vieler Variabeln
in der Theorie der Dirichletschen Reihen Σan /ns , Nachr. Akad. Wiss.
Göttingen math.-Phys. Kl. (1913), 441–488.
[Boh13b] , Über die gleichmässige Konvergenz Dirichletscher Reihen, J.
Reine Angew. Math. 143 (1913), 203–211.
[Car22] F. Carlson, Contributions á la théorie des séries de Dirichlet, Note I,
Ark. Mat. 16 (1922), no. 18, 1–19.
[Car58] L. Carleson, An interpolation problem for bounded analytic functions,
Amer. J. Math. 80 (1958), 921–930.
[CFS82] I. P. Cornfeld, S. V. Fomin, and Ya. G. Sinai, Ergodic theory,
Grundlehren der Mathematischen Wissenschaften [Fundamental Prin-
ciples of Mathematical Sciences], vol. 245, Springer-Verlag, New York,
1982, Translated from the Russian by A. B. Sosinskii.
[DFOC+ 11] A. Defant, L. Frerick, J. Ortega-Cerdà, M. Ounaies, and K. Seip, The
Bohnenblust-Hille inequality for homogeneous polynomials is hypercon-
tractive, Ann. of Math. (2) 174 (2011), no. 1, 485–497.
121
122 BIBLIOGRAPHY
[DS04] Peter Duren and Alexander Schuster, Bergman spaces, Mathematical

Surveys and Monographs, vol. 100, American Mathematical Society,
Providence, RI, 2004.
[Dur70] P. L. Duren, Theory of H p spaces, Academic Press, New York, 1970.
[EFKMR14] Omar El-Fallah, Karim Kellay, Javad Mashreghi, and Thomas Rans-
ford, A primer on the Dirichlet space, Cambridge Tracts in Mathematics,
vol. 203, Cambridge University Press, Cambridge, 2014.
[Fol99] G.B. Folland, Real analysis: Modern techniques and their applications,
Wiley, New York, 1999.
[Gam01] T.W. Gamelin, Complex analysis, Springer, New York, 2001.
[GH99] J. Gordon and H. Hedenmalm, The composition operators on the space
of Dirichlet series with square summable coefficients, Michigan Math. J.
46 (1999), 313–329.
[Hel69] H. Helson, Compact groups and Dirichlet series, Ark. Mat. 8 (1969),
139–143.
[Hel05] , Dirichlet series, Regent Press, Oakland, 2005.
[HKZ00] Haakan Hedenmalm, Boris Korenblum, and Kehe Zhu, Theory of
Bergman spaces, Graduate Texts in Mathematics, vol. 199, Springer-
Verlag, New York, 2000.
[HLS97] H. Hedenmalm, P. Lindqvist, and K. Seip, A Hilbert space of Dirichlet
series and systems of dilated functions in L2 (0, 1), Duke Math. J. 86
(1997), 1–37.
[Kah85] J.-P. Kahane, Some random series of functions, Cambridge University
Press, Cambridge, 1985.
[Koo80] P. Koosis, An introduction to H p , London Mathematical Society Lecture
Notes, vol. 40, Cambridge University Press, Cambridge, 1980.
[McCa04] J.E. McCarthy, Hilbert spaces of Dirichlet series and their multipliers,
Trans. Amer. Math. Soc. 356 (2004), no. 3.
[McCu92] S.A. McCullough, Carathéodory interpolation kernels, Integral Equations
and Operator Theory 15 (1992), no. 1, 43–71.
[McCu94] , The local de Branges-Rovnyak construction and complete
Nevanlinna-Pick kernels, Algebraic methods in Operator theory,
Birkhäuser, 1994, pp. 15–24.
[Mon94] Hugh L. Montgomery, Ten lectures on the interface between analytic
number theory and harmonic analysis, CBMS Regional Conference Se-
ries in Mathematics, vol. 84, Published for the Conference Board of the
Mathematical Sciences, Washington, DC; by the American Mathemati-
cal Society, Providence, RI, 1994.
[MS94] D. Marshall and C. Sundberg, Interpolating sequences
for the multipliers of the Dirichlet space, Preprint; see
http://www.math.washington.edu/∼marshall/preprints/preprints.html,
1994.
[MSS15] Adam W. Marcus, Daniel A. Spielman, and Nikhil Srivastava, Interlac-
ing families II: Mixed characteristic polynomials and the Kadison-Singer
problem, Ann. of Math. (2) 182 (2015), no. 1, 327–350. MR 3374963
[MV] H. Montgomery and U. Vorhauer, Some unsolved problems in har-
monic analysis as found in analytic number theory, http://www.nato-
us.org/analysis2000/papers/montgomery-problems.pdf.
BIBLIOGRAPHY 123
[Nik85] N. K. Nikol’skiı̆, Treatise on the shift operator: Spectral function theory,

Grundlehren der mathematischen Wissenschaften, vol. 273, Springer-
Verlag, Berlin, 1985.
[Ols11] Jan-Fredrik Olsen, Local properties of Hilbert spaces of Dirichlet series,
J. Funct. Anal. 261 (2011), no. 9, 2669–2696.
[OS08] Jan-Fredrik Olsen and Kristian Seip, Local interpolation in Hilbert spaces
of Dirichlet series, Proc. Amer. Math. Soc. 136 (2008), no. 1, 203–212
(electronic).
[QQ13] Hervé Queffélec and Martine Queffélec, Diophantine approximation
and Dirichlet series, Harish-Chandra Research Institute Lecture Notes,
vol. 2, Hindustan Book Agency, New Delhi, 2013.
[Qui93] P. Quiggin, For which reproducing kernel Hilbert spaces is Pick’s theorem
true?, Integral Equations and Operator Theory 16 (1993), no. 2, 244–
266.
[Qui94] , Generalisations of Pick’s theorem to reproducing kernel Hilbert
spaces, Ph.D. thesis, Lancaster University, 1994.
[Rud86] W. Rudin, Real and complex analysis, McGraw-Hill, New York, 1986.
[Sei04] K. Seip, Interpolation and sampling in spaces of analytic functions,
American Mathematical Society, Providence, RI, 2004, University Lec-
ture Series.
[Sei09] Kristian Seip, Interpolation by Dirichlet series in H ∞ , Linear and com-
plex analysis, Amer. Math. Soc. Transl. Ser. 2, vol. 226, Amer. Math.
Soc., Providence, RI, 2009, pp. 153–164.
[Sha87] Joel H. Shapiro, The essential norm of a composition operator, Ann. of
Math. (2) 125 (1987), no. 2, 375–404.
[SS61] H.S. Shapiro and A.L. Shields, On some interpolation problems for an-
alytic functions, Amer. J. Math. 83 (1961), 513–532.
[SS03] E.M. Stein and R. Shakarchi, Fourier analysis, Princeton University
Press, Princeton, 2003.
[SS09] Eero Saksman and Kristian Seip, Integral means and boundary limits of
Dirichlet series, Bull. Lond. Math. Soc. 41 (2009), no. 3, 411–422.
[Tit37] E.C. Titchmarsh, Introduction to the theory of Fourier integrals, Claren-
don Press, Oxford, 1937.
[Tit86] , The theory of the Riemann Zeta-function, second edition, Ox-
ford University Press, Oxford, 1986.
Index
Page numbers appear in bold for pages where the term

is defined.
E(ε, f ), 63 T∞ , 43
H∞ p
(Ω1/2 ), 97 Tr , 45
H∞ p
(Ω1/2 , 97 Hw2 , 67, 69
2
K, 77 Hw , 66, 69, 75
QK , 72 d(k), 8
B, 42 dj (k), 8
Xq , 87 gi , 105
D, 69 mte -Abschnitt, 62
E, 105 z r(n) , 42
H2 , interpolating sequences, 107
2 `2 (G), 87
Hw , interpolating sequences, 107
H∞ , 101
Hp , 97
σb , 38
Mult (H), 118
σu , 38
Q, 42
uniformly local H p space on Ω1/2 , 97
Θ(x), 12
ζ̃(s): alternating zeta function, 25 Abel’s Summation by parts formula,
β, 55 4
E, 44 abscissa of absolute convergence, 25,
NK , 72 27
ε-translation number, 63, 64 abscissa of bounded convergence, σb ,
RT
−−T
, 33 38
Ωρ , 3 abscissa of convergence, 25, 26
H2 , 57, 69 abscissa of uniform convergence, σu ,
.: less than or equal to a constant 38
times, 27 alternating zeta function, 25, 37
µ(n), 6
µα , 66 Beurling’s question, 56, 74
π(x), 5 Birkhoff-Khinchin Theorem, 77
Bohr condition, 23
P, 2
Pk , 2 Carleson measure, 101
ρ, 77 Carlson’s theorem, 68, 75, 78
ρA , 100 character, 62
σa : abscissa of absolute cyclic vector, 74
convergence, 25
σc : abscissa of convergence, 25 Dirichlet character modulo q, 87
125
126 INDEX
Dirichlet L-function, 88 Schnee’s theorem, 33, 35

dual group, 87, 93 strongly separated, 100
dual system, 105 summatory function, 31
dual system, minimal, 105
three line lemma, 70
ergodic, 77 topologically free, 105
ergodic theorem, 77
uniformly almost periodic, 63
Euler product formula, 2, 10, 88
Euler totient function, 9 vertical limit function, 79
von Mangoldt function, 14
flow, 77
frame, 56 weakly separated, 100
function representable by Dirichlet
series, 32
generalized Dirichlet series, 23, 28,

35
Gleason distance, 100
Gram matrix, 55, 106
infinite polydisk, 62
interpolating sequence, for a Hilbert
function space, 105
interpolation constant, 99
Inverse Fourier Transform, 29
Inverse Mellin Transform, 30
kernel function, 117

Kronecker flow, 77
Kronecker’s theorem, 60
Möbius function, µ(n), 6

Möbius inversion formula, 9
Mellin transform, 29
multiplier algebra, 57, 118
Perron’s formula, 31, 32, 35

Pick property, 107
Prime number theorem, 12
prime zeta function, 108
pseudo-hyperbolic metric, 100
quasi-character, 10, 62
Rademacher sequence, 44
relatively dense, 63
reproducing kernel, 117
Riemann hypothesis, 1
Riemann zeta function, 1
Riesz basis, 55

Dirichlet Series

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dirichlet Series

Uploaded by

Copyright:

Available Formats

Dirichlet Series

7.1. Vertical Limits 77

In 2005, I taught a graduate course on Dirichlet series at Washing-

A Dirichlet series is a series of the form

The most famous example is the Riemann zeta function

Notation 1.1. By long-standing tradition, the complex variable

The importance of the Riemann zeta function and the Riemann

Theorem 1.5. (Euler Product formula) For σ > 1,

Y −1  −1  −1  −1

If we formally multiply out this infinite product, we can only obtain

For a rigorous proof assume that Re s > 1, and fix k ∈ N+ . Then

where the last equality holds by a variation of the formal argument

Thus the product converges to ζ(s).

Notation 1.7. We shall let Ωρ denote the open half-plane

Proof: Suppose not, then p∈P p1 converges. By the Taylor expan-

sion of log(1 − x), for x close enough to 0,

PnLemma 1.10. (Abel’s Summation by parts formula) Let An =

Proof: Since ak = Ak − Ak−1 , we have

Notation 1.11. For x > 0, we let π(x) denote the number of

Notation 1.13. The Möbius function is helpful when working

Proof: We only present a formal proof — convergence can be

P Proof: As remarked, we may assume that ρ < 0, so in particular,

the claim is proved.

if at least one of the sums on the left-hand side converges absolutely.

with absolute convergence.

Corollary 1.17. For σ > 1,

where d(k) denotes the number of divisors of k. More generally,

The proof of the second formula is analogous.

Comparing the coefficients of the outer-most Dirichlet series completes

Proposition 1.19. (Möbius inversion formula) Let f, g be

Definition 1.20. The Euler totient function φ(n) is defined as

Theorem 1.22. For σ > 2,

Proof: Again, we will only prove it formally, since turning it into a

The Prime Number Theorem

2.1. Statement of the Prime number theorem

so π(n) cannot be O(nα ) for any α < 1.

exists, it must be 1. The full prime number theorem was proved in

Looking at some examples, we see that

Definition 2.3. For s ∈ Ω1 , we define

It is easy to see that this Dirichlet series converges absolutely in Ω1 .

The key to proving the Prime number theorem is establishing the

2.2. Proof of the Prime number theorem

Now, by the binomial theorem,

Since Θ(1) = 0, we conclude that

By logarithmic differentiation we obtain

The second term on the RHS defines an analytic function in Ω1/2 as

Proposition 2.15. For s ∈ Ω1

−ζ 0 (s) X Λ(n) X log p

This double summation goes over exactly those numbers n = pk for

Lemma 2.18. Let x0 ∈ R and assume F is holomorphic in a neigh-

for x ∈ (x0 − ε, x0 + ε).

Proof: Write F (x) = ak (s − x0 )k + ak+1 (s − x0 )k+1 + . . . where

F 0 (x) ζ 0 (x) ζ 0 (x + it0 ) ζ 0 (x + 2it0 )

We observe that −Λ(n)n−x ≤ 0 for every n ≥ 2 while the term in the

Then gT is an entire function by Morera’s theorem, and gT (0) =

For s ∈ C+ , we have |s| = R and so

and the expression in square bracket is independent of T and holo-

and thus g(s) is the Laplace

Note 2.26. See [Fol99, p. 107] for information on integration by

Proof: We will proceed by contraction. There are two cases.

Proof of Theorem 2.2. We can estimate

= [π(x) − π(x1−ε )](1 − ε) log x

2.3. Historical Notes

Convergence of Dirichlet Series

Y −1 −1 −1 −1

where CN is the number of terms, roughly N r , if N r.